Method for detecting ablation degree of water-blocking buffer layer of cable

Abstract

The invention provides a cable water-blocking buffer layer ablation degree detection method. The method comprises the following steps: acquiring a visible light image and an infrared image of a water-blocking buffer layer of a cable; performing weighted fusion on the visible light image and the infrared image to obtain a target image; performing multi-scale decomposition on the target image to obtain a plurality of spatial frequency images corresponding to the target image, the multi-scale decomposition being decomposition of the image into different spatial frequencies; and under the condition that the spatial frequency corresponding to the at least one spatial frequency image is greater than a preset frequency threshold, determining the ablation of the water-blocking buffer layer of the cable, and determining the ablation degree of the water-blocking buffer layer of the cable according to the target image. According to the scheme, the problem that the ablation degree cannot be accurately evaluated by adopting single-dimensional data in the prior art is solved.

Description

Technical Field

[0001] This application relates to the field of cable testing technology, and more specifically, to a method for detecting the degree of erosion of the water-blocking buffer layer of a cable, a device for detecting the degree of erosion of the water-blocking buffer layer of a cable, a computer program product, and a cable testing system. Background Technology

[0002] The water-blocking buffer layer of cables in power systems is a key structure to ensure the long-term stable operation of cables, and its integrity directly affects the safety and stability of power transmission.

[0003] Traditional cable water-blocking buffer layer erosion detection mainly relies on manual inspection and simple instrument measurements, such as using infrared thermal imagers to detect abnormal surface temperatures, using ultrasonic detectors to check for physical damage, or using electrical parameter measurement methods to evaluate insulation performance.

[0004] However, these methods typically only obtain detection data from a single dimension, making it impossible to accurately assess the degree of ablation. Summary of the Invention

[0005] The main objective of this application is to provide a method, device, computer program product, and cable testing system for detecting the degree of erosion of a cable water-blocking buffer layer, so as to at least solve the problem that the existing technology cannot accurately assess the degree of erosion using data from a single dimension.

[0006] To achieve the above objectives, according to one aspect of this application, a method for detecting the degree of erosion of a cable's water-blocking buffer layer is provided, comprising: acquiring a visible light image and an infrared image of the cable's water-blocking buffer layer; weightedly fusing the visible light image and the infrared image to obtain a target image; performing multi-scale decomposition on the target image to obtain multiple spatial frequency images corresponding to the target image, wherein the multi-scale decomposition involves decomposing the image into different spatial frequencies; determining that the water-blocking buffer layer of the cable is eroded when at least one of the spatial frequency images corresponds to a spatial frequency greater than a preset frequency threshold, and determining the degree of erosion of the water-blocking buffer layer of the cable based on the target image.

[0007] According to another aspect of this application, a device for detecting the degree of erosion of a cable's water-blocking buffer layer is provided, comprising: a first acquisition unit for acquiring a visible light image and an infrared image of the cable's water-blocking buffer layer; a first fusion unit for weighted fusion of the visible light image and the infrared image to obtain a target image; a decomposition unit for performing multi-scale decomposition on the target image to obtain multiple spatial frequency images corresponding to the target image, wherein the multi-scale decomposition is to decompose the image into different spatial frequencies; and a first determination unit for determining the erosion of the cable's water-blocking buffer layer and determining the degree of erosion of the cable's water-blocking buffer layer when the spatial frequency corresponding to at least one of the spatial frequency images is greater than a preset frequency threshold.

[0008] According to another aspect of this application, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of any of the cable water-blocking buffer layer ablation degree detection methods.

[0009] According to another aspect of this application, a cable inspection system is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including methods for performing any of the cable water-blocking buffer layer ablation detection methods described above.

[0010] The technical solution of this application can effectively display physical damage and material changes on the surface using visible light images, but it has low sensitivity to temperature distribution and thermal damage. Infrared images can accurately capture temperature anomalies, but they lack detailed description of surface texture and structural damage. Therefore, this solution combines the characteristics of these two images and fuses them together, thereby ensuring that the target image has the characteristics of both images. This provides a more comprehensive data foundation for subsequent ablation degree detection. Attached Figure Description

[0011] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0012] Figure 1 A hardware structure block diagram of a mobile terminal for performing a method for detecting the degree of ablation of a cable water-blocking buffer layer is shown in an embodiment of this application.

[0013] Figure 2 A schematic flowchart of a method for detecting the degree of erosion of a cable water-blocking buffer layer according to an embodiment of this application is shown.

[0014] Figure 3 A flowchart illustrating another method for detecting the degree of erosion of the water-blocking buffer layer in cables is shown.

[0015] Figure 4 A structural block diagram of a cable water-blocking buffer layer ablation detection device provided according to an embodiment of this application is shown.

[0016] The above figures include the following reference numerals:

[0017] 102. Processor; 104. Memory; 106. Transmission device; 108. Input / output device. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0020] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] Existing automated detection technologies, such as video image recognition systems and fixed infrared monitoring systems, while improving detection efficiency, still have significant limitations in extracting ablation features and assessing risks. In recent years, with the development of technologies such as multispectral imaging and deep learning, some research has begun to explore the application of multi-source data fusion and intelligent analysis to cable condition monitoring, but a complete detection methodology system has not yet been formed. Single sensing methods are insufficient to comprehensively capture the multidimensional features of ablation, leading to incomplete detection results.

[0022] As described in the background section, existing technologies using single-dimensional data cannot accurately assess the degree of ablation. To address the above problem, embodiments of this application provide a method for detecting the degree of ablation of a cable water-blocking buffer layer, a device for detecting the degree of ablation of a cable water-blocking buffer layer, a computer program product, and a cable testing system.

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0024] The methods and embodiments provided in this application can be executed on a mobile terminal, computer terminal, or similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for a method of detecting the degree of erosion of a cable water-blocking buffer layer according to an embodiment of the present invention. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. The mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0025] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the cable water-blocking buffer layer erosion detection method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the aforementioned networks may include wireless networks provided by the mobile terminal's communication provider. In one example, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0026] This embodiment provides a method for detecting the degree of erosion of the water-blocking buffer layer of a cable running on a mobile terminal, computer terminal or similar computing device. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Although the logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0027] Figure 2 This is a schematic flowchart of a method for detecting the degree of erosion of a cable water-blocking buffer layer according to an embodiment of this application. Figure 2 As shown, the method includes the following steps:

[0028] Step S201: Obtain visible light and infrared images of the water-blocking buffer layer of the cable;

[0029] Specifically, acquiring visible light and infrared images of the cable's water-blocking buffer layer is achieved using a dual-spectrum imaging device. This device is equipped with both a visible light camera and an infrared camera, enabling simultaneous acquisition of visible light and infrared spectral information from the cable surface within the same field of view. For example, the visible light camera captures images in the 400-700 nm wavelength range, while the infrared camera captures thermal distribution information in the 3-5 μm wavelength range. To ensure image accuracy and quality, the acquisition device must operate under suitable lighting conditions to avoid interference from ambient light on the infrared images.

[0030] Step S202: The visible light image and the infrared image are weighted and fused to obtain the target image;

[0031] Specifically, after acquiring visible light and infrared images of the cable's water-blocking buffer layer, these two types of images are weighted and fused to generate a fused target image. The fusion process employs wavelet transform for multi-scale analysis, decomposing the visible light and infrared images into detail and approximate components across multiple frequency bands. Then, these components are weighted and summed based on the importance of information in each frequency band to obtain the fused image. The weighting coefficients are calculated based on the image's local entropy and gradient magnitude to automatically adjust the fusion weights for different frequency bands.

[0032] Step S203: Perform multi-scale decomposition on the target image to obtain multiple spatial frequency images corresponding to the target image, wherein the multi-scale decomposition is to decompose the image into different spatial frequencies.

[0033] Specifically, the weighted and fused target image is decomposed into multi-scale components to obtain image components at different spatial frequencies. Multi-scale decomposition is typically achieved using wavelet transform, which decomposes the target image into low-frequency approximate components and a series of high-frequency detail components. Each high-frequency detail component corresponds to a different spatial frequency, revealing the texture and ablation characteristics of the cable surface at different scales.

[0034] Step S204: If the spatial frequency corresponding to at least one of the above spatial frequency images is greater than a preset frequency threshold, it is determined that the water-blocking buffer layer of the cable is ablated, and the degree of ablation of the water-blocking buffer layer of the cable is determined according to the target image.

[0035] Specifically, the decomposed spatial frequency images are examined. When at least one spatial frequency image is found to have a frequency greater than a preset frequency threshold, it is determined that the cable's water-blocking buffer layer has been ablated. The preset frequency threshold is used to distinguish between normal textures and ablation boundary features and can be adjusted according to the cable material characteristics and actual needs. Once ablation is confirmed, the size, shape, and depth changes of the ablated area in the target image are further analyzed to comprehensively assess the severity of the ablation and generate a graded assessment report.

[0036] In this embodiment, visible light images can better display the physical damage and material changes of the surface, but have low sensitivity to temperature distribution and thermal damage. Infrared images can accurately capture temperature anomalies, but lack detailed description of surface texture and structural damage. Therefore, this solution combines the characteristics of these two images and fuses them together, thereby ensuring that the target image has the characteristics of both images. This provides a more comprehensive data foundation for subsequent ablation degree detection.

[0037] Specifically, such as Figure 3 As shown, this solution mainly includes the following steps:

[0038] 1. A dual-spectrum imaging device was used to simultaneously acquire visible and infrared light data of the water-blocking buffer layer of the cable. After nonlinear contrast optimization processing, an enhanced dual-spectrum image was obtained.

[0039] 2. The enhanced bispectral image is used to locate the ablation region by combining adaptive region growing segmentation with edge preserving filtering, and the texture gradient feature matrix is ​​extracted to obtain the ablation feature spectrum (i.e., image features).

[0040] 3. Based on the above ablation feature spectrum, an ablation degree assessment model is established using an improved residual attention network. The model is then trained through transfer learning to form an ablation grading engine (i.e., a degree recognition model).

[0041] 4. Based on the output results of the ablation grading engine, the micro-morphology of the ablation surface is constructed using phase structured light coding technology to generate a high-precision ablation depth map (i.e., ablation image).

[0042] 5. The above ablation depth map and the above ablation classification engine results are integrated through a self-correcting weighted fusion algorithm to calculate the ablation risk quantification index (i.e. the current operational risk of the cable);

[0043] 6. Based on the above-mentioned quantitative indicators of ablation risk, a time-series prediction and extrapolation mechanism is used to generate a distributed visualization report and early warning threshold suggestions.

[0044] Specifically, the executing entity of this application can be a cable detection system for assessing the degree of erosion of the cable's water-blocking buffer layer, or it can be a terminal or a server; no specific limitation is made here. This application's embodiment uses a server as an example for illustration.

[0045] Specifically, a dual-spectrum imaging device was used to simultaneously acquire visible and infrared light data of the cable's water-blocking buffer layer. The dual-spectrum imaging device consists of a visible light camera and an infrared camera, which are simultaneously imaged in the same target area using an optical beam splitter. In actual testing, the surface of the cable's water-blocking buffer layer undergoes ablation due to thermal damage. This ablation often leads to changes in the material's physical properties, resulting in different spectral response characteristics at different wavelengths. After simultaneously acquiring images of the cable surface using a dual-sensor array to form raw dual-channel data, a polarization filtering matrix is ​​applied to this data to eliminate surface reflection interference and extract the spectral response signal unique to ablation. The polarization filtering matrix effectively eliminates specular reflection from the cable surface and enhances the spectral characteristics of the material itself by adjusting the combination of polarization filters at different angles. Subsequently, the spectral response signal is decomposed into multiple scales using wavelet transform to separate the thermal damage and mechanical damage information layers. The differences in signal intensity distribution at different frequency bands distinguish between thermal ablation and mechanical damage. A local frequency domain enhancement algorithm is applied to the thermal and mechanical damage information layers to highlight ablation boundary features. Finally, the spectral response signal and ablation boundary features are integrated through an adaptive weighted fusion processor to generate an enhanced bispectral image with the ability to distinguish between shallow and deep layers. When locating ablation regions using an enhanced bispectral image through a combination of adaptive region growing segmentation and edge-preserving filtering, variational multi-scale decomposition is first performed on the enhanced bispectral image to separate the ablation feature frequency domain from the background frequency domain. Variational multi-scale decomposition, based on the image's brightness, texture, and structural information, decomposes the image into components of different scales, where the high-frequency part contains ablation feature details and the low-frequency part contains background information. Next, the ablation feature frequency domain is used to determine initial seed points through an adaptive gradient threshold to establish a multi-core growth region map. The adaptive gradient threshold dynamically adjusts the threshold according to the gray-level gradient distribution of the local region, selecting seed points at the edges of the ablation region. A tensor projection algorithm is used to calculate the principal direction vector field for the multi-core growth region map, constructing ablation boundary constraints. The tensor projection algorithm determines the principal direction of the texture by calculating the eigenvalues ​​and eigenvectors of the local image structure tensor, forming a direction vector field. By combining ablation boundary constraints with a multi-core growth region map, anisotropic diffusion filtering is used to suppress false boundaries, forming an accurate ablation region contour. From this accurate contour, gray-level co-occurrence matrix (GLCM), local binary patterns (LANDs), and higher-order moment statistics are extracted to construct an ablation texture feature space. The GLCM describes the spatial correlation of image texture, LANDs capture local texture patterns, and higher-order moment statistics characterize pixel distribution characteristics. Principal component analysis (PCA) is used to reduce the dimensionality of the ablation texture feature space, and feature selection is performed to generate an ablation feature spectrum representing the ablation depth.

[0046] Specifically, when establishing an ablation degree assessment model using an improved residual attention network based on the ablation feature spectrum, the ablation feature spectrum is first enhanced in the spectral domain by separating ablation-specific frequency band information through multi-kernel orthogonal transformation. Multi-kernel orthogonal transformation maps frequency domain information through different kernel functions, highlighting the ablation-specific frequency band response. The ablation-specific frequency band information is then imported into a bidirectional gated feature selection unit to extract a discriminative feature set. This unit, through feature flow in both forward and reverse directions combined with a gating mechanism, controls information transmission and extracts the most discriminative features. A multi-layer residual block structure is constructed for the discriminative feature set, with a channel attention mechanism introduced in each layer to form an adaptive feature representation. The residual block structure addresses the gradient vanishing problem in deep network training through skip connections, while the channel attention mechanism adaptively adjusts the weights of different channels, strengthening ablation-related features. The adaptive feature representation is then combined with a pre-trained knowledge base for transfer weight mapping, generating an ablation knowledge prior model. This transfer weight mapping accelerates model convergence by extracting general feature representation capabilities from the pre-trained model and transferring them to the ablation recognition task. The ablation knowledge prior model is optimized for parameters using a cross-validation fine-tuning strategy to establish ablation category distinction boundaries. These boundaries are then combined with ablation-specific frequency band information to construct an interpretable inference path, forming an ablation grading engine. When constructing the microstructure of the ablation surface based on the engine's output and phase structured light coding technology, the engine's output is used as a region-guided constraint to mark key detection zones. A composite sinusoidal phase-coded grating is projected onto these key detection zones, generating a variable-density fringe projection pattern. This composite sinusoidal phase-coded grating, through the combination of multi-frequency, multi-phase sinusoidal fringes, forms a unique coding pattern for high-precision 3D reconstruction. Multi-angle reflection responses are captured from the variable-density fringe projection pattern, and phase information is extracted using Fourier transform. The multi-angle reflection responses are captured by cameras from different viewpoints to observe fringe deformation, and Fourier transform is used to extract phase information from the deformed fringes. The phase information is then processed using a Gauss-Helmholtz phase unwrapping algorithm to eliminate phase jumps and construct a continuous depth map. The Gauss-Helmholtz phase unwrapping algorithm transforms the wrapped phase into a continuous phase by solving partial differential equations, avoiding depth discontinuities caused by phase jumps. Spatial registration is performed between the continuous depth map and the key ablation detection region, and the Laplacian edge enhancement algorithm highlights the micro-texture. The micro-texture is then fused with the continuous depth map, and anisotropic filtering is used to preserve structural boundaries, generating a high-precision ablation depth map.

[0047] Specifically, when calculating the ablation risk quantification index by integrating the ablation depth map and the ablation grading engine results using a self-calibrating weighted fusion algorithm, the first step is to extract the depth ripple spectrum, material embrittlement region map, and microcrack distribution map from the ablation depth map to construct a three-dimensional damage feature map. The depth ripple spectrum reflects surface undulations, the material embrittlement region map represents areas of decreased material strength, and the microcrack distribution map shows the crack network distribution. Heterogeneous data registration is then performed between the three-dimensional damage feature map and the ablation grading engine results to establish a multi-level ablation characterization space. Heterogeneous data registration aligns data from different sources and dimensions to a unified coordinate system through feature point matching and spatial transformation. Tensor decomposition technology is applied to the multi-level ablation characterization space to separate thermal and mechanical damage modes, forming a dual-modal damage component. The dual-modal damage component is processed through a nonlinear self-calibrating network, and parameters are dynamically adjusted based on a historical case library to generate an optimized weight vector. Based on the optimized weight vector, the three-dimensional damage feature map and the ablation grading engine results are adaptively fused to construct an ablation damage integrity assessment system. The severity of the current ablation state is calculated based on the ablation damage integrity assessment system, and standardized ablation risk quantification indicators are output. Using a time-series prediction and extrapolation mechanism based on these indicators, a distributed visualization report and early warning threshold suggestions are generated. The ablation risk quantification indicators are then decomposed over time to separate long-term trends, periodic changes, and random disturbances, constructing an ablation evolution pattern library. This library is input into a recursive waveform recognizer to extract typical degradation path features and establish an ablation propagation prediction baseline. Multi-scale entropy analysis is applied to this baseline to determine the critical transition points between the stable zone, accelerated degradation zone, and danger zone, forming an ablation early warning threshold spectrum. This spectrum is converted into a color-coded interactive layer and overlaid on a 3D model of the cable structure to generate a three-dimensional visualization interface. A timeline control component is added to this interface to enable ablation state retrospection and forward-looking extrapolation, constructing an early warning decision support system. The data analysis and prediction results generated by the early warning decision support system are integrated into a standardized report format, outputting a distributed visualization report and early warning threshold suggestions.

[0048] For example, when detecting ablation of the water-blocking buffer layer of a high-voltage cable in a power transmission network, a dual-spectrum imaging device acquired image data in the visible light band (400-700nm) and the mid-infrared band (3-5μm). Surface reflection interference was eliminated using a polarization filtering matrix, and the unique spectral response curve of the ablated region was extracted. A significant absorption peak was found at 4.3μm, indicating thermal decomposition of the organic material. Wavelet transform decomposed the signal into four scale layers, with the second and third layers containing significant ablation feature information. An adaptive region growing algorithm started from the severely thermally damaged core region and expanded outwards to form a complete ablation region outline. Texture features extracted from the ablated region included the contrast, correlation, energy, and entropy values ​​of the gray-level co-occurrence matrix, as well as the uniformity distribution characteristics of local binary patterns. An improved residual attention network transferred weights from a pre-trained model to the ablation recognition task through transfer learning, and fine-tuned the network parameters through five-fold cross-validation, ultimately achieving accurate classification of the ablation degree. Phase-structured optical coding technology projects 16-phase-shifted sinusoidal fringes into the ablation region, and reconstructs the three-dimensional surface morphology using a phase unwrapping algorithm to accurately measure the ablation depth distribution. Multimodal fusion technology combines depth information with material property information to calculate a comprehensive risk index and predicts the ablation propagation trend based on historical data, providing a basis for maintenance decisions.

[0049] In the specific implementation process, the degree of ablation of the water-blocking buffer layer of the cable can be determined based on the target image by the following steps: extracting features from the target image to obtain image features; forming a training set by combining historical image features and corresponding degree labels, and training an image recognition model using the training set to obtain a degree recognition model, wherein the degree labels are the historical ablation degrees of the historical image features in the training set; and inputting the image features into the degree recognition model to obtain the ablation degree corresponding to the image features.

[0050] In this scheme, the deep learning-based ablation severity recognition model, trained on a large amount of historical data, can automatically learn the complex mapping relationship between image features and ablation severity, enabling accurate identification even for minor ablation with inconspicuous features or severe ablation with complex structures. The construction of the training set and the setting of labels ensure the model's generalization ability and prediction accuracy. The diversity of historical image features helps the model learn feature representations under different ablation patterns, while accurate ablation severity labels enable the model to accurately classify the ablation severity, thus ensuring accurate assessment of the ablation severity.

[0051] In some embodiments, the extraction of features from the target image to obtain image features can be achieved through the following steps: extracting at least one spatial frequency image corresponding to the target image whose spatial frequency is greater than the preset frequency threshold, and determining the portion of the spatial frequency image whose spatial frequency is greater than the preset frequency threshold in the target image as a preliminary ablation region; extracting the center point or spatial frequency peak point of the preliminary ablation region in the target image to obtain seed points; controlling the growth of the seed points in the preliminary ablation region and smoothing them using anisotropic diffusion filtering technology to obtain the ablation region; extracting texture features from the ablation region and performing dimensionality reduction using principal component analysis to obtain the image features.

[0052] In this scheme, multi-scale decomposition can decompose the image according to different frequencies, thereby capturing the ablation features at different scales. The selection of seed points is based on the center point or frequency peak point of the initial ablation region, which can guide the region growing algorithm to more accurately segment the ablation region, rather than treating the entire cable surface as the processing object. The selection of seed points and the application of the region growing algorithm realize the automatic expansion of the ablation region. Even under complex backgrounds and noise interference, a clear and complete ablation region image can be obtained. The smoothing processing of anisotropic diffusion filtering technology effectively suppresses image noise and maintains the sharpness of the ablation region boundary. The dimensionality reduction processing of principal component analysis identifies the main change directions in the original feature vector through statistical analysis. These directions are used as the feature axes after dimensionality reduction, which significantly reduces the dimensionality of the feature vector. At the same time, redundant information and noise interference are removed, thereby ensuring that the quality of the extracted image features is good.

[0053] Current solutions lack dedicated image processing and feature extraction algorithms for the characteristics of water-blocking buffer layers, reducing detection accuracy. Specifically, the enhanced bispectral image is used to locate the ablation region by combining adaptive region growing segmentation with edge-preserving filtering, extracting the texture gradient feature matrix to obtain the ablation feature spectrum. This includes: performing variational multi-scale decomposition on the enhanced bispectral image to separate the ablation feature frequency domain from the background frequency domain; determining initial seed points for the ablation feature frequency domain using adaptive gradient thresholding to establish a multi-core growth region map; calculating the principal direction vector field for the multi-core growth region map using tensor projection algorithm to construct ablation boundary constraints; combining the ablation boundary constraints with the multi-core growth region map, and suppressing false boundaries through anisotropic diffusion filtering to form a precise ablation region contour; extracting the gray-level co-occurrence matrix, local binary mode, and higher-order moment statistics from the precise ablation region contour to construct an ablation texture feature space; and using principal component analysis for dimensionality reduction and feature selection to generate an ablation feature spectrum (i.e., image features) characterizing the ablation depth.

[0054] Specifically, variational multi-scale decomposition (VMD) is performed on the enhanced bispectral image. This step effectively distinguishes ablation features from the background by separating different frequency domain components. VMD is a signal processing technique that decomposes an image into different frequency components at multiple scales. In the detection of ablation of the cable water-blocking buffer layer, VMD decomposes the enhanced bispectral image into three frequency domain components: high frequency, mid frequency, and low frequency. The high-frequency component mainly contains ablation edge and texture feature information, the mid-frequency component contains the main feature information of the ablation area, and the low-frequency component mainly contains background information. By analyzing the energy distribution characteristics of each frequency domain component, the frequency domain of ablation features can be effectively distinguished from the frequency domain of the background.

[0055] Variational multiscale decomposition divides the image into three main frequency domain components: high-frequency, mid-frequency, and low-frequency. The ablation feature frequency domain is mainly concentrated in the high and mid-frequency bands: high-frequency components correspond to changes in surface texture, edge details, and microstructural damage caused by ablation, with a frequency range of approximately ten cycles per millimeter or more; mid-frequency components correspond to the main features of the ablated region, including macroscopic features such as material discoloration and thickness changes, with a frequency range between one and ten cycles per millimeter. The background frequency domain is mainly composed of low-frequency components, corresponding to the uniform material properties and slowly changing environmental factors in undamaged areas, with frequencies below one cycle per millimeter. This separation is not based on a fixed numerical threshold but is dynamically determined through energy distribution analysis: the system calculates the energy proportion of each frequency domain component; when the proportion of high-frequency and mid-frequency energy to the total energy exceeds a set threshold, the region is classified as the ablation feature frequency domain; when low-frequency energy dominates, it is classified as the background frequency domain. This dynamic separation method can adapt to different degrees of ablation and different material properties.

[0056] After separating the ablation feature frequency domain, initial seed points are determined using an adaptive gradient threshold to establish a multi-core growth region map. The adaptive gradient threshold technique dynamically adjusts the threshold based on local gradient changes in the image, adapting to image features with varying degrees of ablation. The selection of initial seed points directly affects the accuracy of subsequent region growth; seed points must be located within the ablation region and be representative. The system first calculates the gradient magnitude map of the image, then automatically determines a suitable gradient threshold based on the statistical characteristics of the ablation feature frequency domain, selecting points with gradient values ​​greater than the threshold as initial seed points. When multiple discontinuous ablation points are detected on the cable surface, multiple seed point sets are formed, thus establishing a multi-core growth region map and providing a foundation for the subsequent region growth process.

[0057] In practical cable ablation detection, ablation is often not a single continuous area, but rather multiple independent or semi-independent damage points distributed across the cable surface. These damage points may be caused by different reasons and have different ablation degrees and development stages. Traditional single-core region growth algorithms can only expand from a single seed point, failing to handle multiple dispersed ablation regions simultaneously, easily leading to missed detections or false connections. The process of establishing a multi-core growth region map is as follows: multiple representative seed points are identified in the ablation feature frequency domain using an adaptive gradient threshold. These seed points are located at the center of each independent ablation region or at a location with obvious features. Then, starting from each seed point, multiple independent region growth processes are initiated simultaneously. Each growth process expands outward according to a similarity criterion until it encounters a clear boundary or meets other growth regions. This multi-core parallel growth method ensures that all ablation regions are accurately identified and segmented, avoiding the omissions that may occur with single-core methods, while maintaining the independence of each region, facilitating subsequent classification analysis and risk assessment. It improves the detection completeness of complex ablation patterns and enhances the algorithm's adaptability to different ablation distributions.

[0058] The tensor projection algorithm is used to calculate the principal direction vector field for the formed multi-core growth region map, constructing ablation boundary constraints. The tensor projection algorithm analyzes the directional information in the image through local structure tensors, accurately capturing the directional characteristics of the ablation boundary. For each pixel, the system constructs its local structure tensor, and calculates the principal direction of that point by analyzing the eigenvalues ​​and eigenvectors of the tensor. The principal direction vector field represents the tangent direction of the ablation region boundary, providing directional guidance for subsequent boundary constraints. When applied to ablation images of cable water-blocking buffer layers, the principal direction vector field accurately reflects the morphological characteristics of the ablation boundary, showing good adaptability, especially for irregularly shaped ablation regions.

[0059] The tensor projection algorithm calculates a local structure tensor for each pixel in the multi-core growth region map. This tensor describes the gradient distribution characteristics around that point. By performing eigenvalue decomposition on the structure tensor, two main eigenvectors are obtained: the first eigenvector points to the direction of the largest local gradient change, typically corresponding to the boundary normal; the second eigenvector points to the direction of the smallest gradient change, corresponding to the boundary tangent. The principal direction vector field is composed of these tangent vectors, describing the preferred direction for the ablation boundary to extend. The establishment of ablation boundary constraints is more complex: the system combines the principal direction vector field with the geometric information of the multi-core growth region to establish directional and continuity constraints for each boundary point. Directional constraints require that the boundary orientation should be consistent with the local principal direction vector, with a deviation angle not exceeding a set threshold; continuity constraints require that the directional changes between adjacent boundary points should transition smoothly, avoiding abrupt changes. These constraints play a guiding role in the subsequent anisotropic diffusion filtering process, ensuring the smoothness and rationality of the boundaries. The technical effects are as follows: it significantly improves the accuracy of boundary positioning, especially for irregularly shaped ablation areas; it effectively suppresses noise interference with boundary detection; and it ensures the geometric continuity and physical rationality of the boundary.

[0060] After combining the ablation boundary constraints with the multi-core growth region map, anisotropic diffusion filtering is used to suppress false boundaries and form an accurate ablation region contour. Anisotropic diffusion filtering is an edge-preserving smoothing technique that smooths along the boundary direction while maintaining boundary sharpness. In cable ablation detection, anisotropic diffusion filtering uses the principal direction vector field obtained in the previous steps as diffusion guidance, performing strong smoothing parallel to the boundary direction while maintaining boundary clarity perpendicular to the boundary direction. This processing method can effectively suppress false boundaries caused by image noise or material inhomogeneity, while enhancing the true ablation boundary. In the processed image, the ablation region boundary becomes clearer and more coherent, laying the foundation for accurate extraction of the ablation region contour.

[0061] Ablation boundary identification is a preliminary step based on differences in spectral characteristics. This boundary is often a collection of discontinuous, noisy fragments, and may contain many spurious boundaries. Contour processing, on the other hand, aims to obtain a continuous, geometrically sound, and complete contour based on the preceding steps. The specific processing steps include: suppressing spurious boundaries through anisotropic diffusion filtering, which smooths along the true boundary direction while maintaining sharpness perpendicular to the boundary, thus eliminating spurious boundaries caused by image noise or material inhomogeneity; applying constraints from the principal direction vector field to guide boundary connection and extension, connecting scattered boundary fragments into a closed contour; and performing boundary optimization processing, including smoothing, deburring, and closure / break operations. The technical value of this contour processing lies in: providing the precise geometric boundaries required for subsequent analysis, ensuring the accuracy of feature extraction; eliminating unavoidable noise and errors in the initial processing; and providing a reliable boundary benchmark for 3D reconstruction and risk assessment. Therefore, this is not repetitive processing, but a necessary refinement process from a coarse boundary to a precise contour.

[0062] Extracting texture features from the precise contours of the ablation region is a crucial step in constructing the ablation texture feature space. The gray-level co-occurrence matrix (GLCM) is a statistical texture feature description method that describes the spatial correlation of pixel gray values ​​in an image. The system calculates GLCMs at different directions and distances, extracting statistical features such as contrast, correlation, energy, and entropy. These features reflect the texture complexity and regularity of the ablation region. Local binary mode (LBM) is an efficient texture descriptor that generates binary codes by comparing the gray values ​​of the central pixel with its neighboring pixels, thereby constructing histogram features. Higher-order moment statistics capture subtle texture changes on the ablation surface by calculating higher-order statistical properties such as skewness and kurtosis of the gray value distribution in the ablation region. These three types of features together constitute a multidimensional ablation texture feature space, comprehensively describing the texture characteristics of the ablation region.

[0063] Three important types of texture feature data are extracted from the ablation region contour. The first type is the gray-level co-occurrence matrix (GLCM) feature, which records the spatial co-occurrence relationship of pixels with different gray values, describing the roughness, regularity, and directionality of the texture. Statistical features such as contrast, correlation, energy, and entropy are extracted from this matrix, each reflecting different physical properties of the ablated surface. The second type is the local binary pattern feature, which captures the microstructure of local textures. By comparing the gray-level relationship between the central pixel and its neighboring pixels to generate binary codes, a histogram feature is constructed. This feature is insensitive to changes in illumination and can stably describe the surface texture changes caused by ablation. The third type is higher-order moment statistics, including skewness and kurtosis. These statistics describe the shape characteristics of the pixel gray-level distribution, reflecting the influence of the ablation degree on the surface reflectivity. These three types of features together constitute a multi-dimensional ablation texture feature space, with each ablated region corresponding to a feature vector in this space. The technical significance of constructing this feature space lies in the following: it provides a quantitative description of the degree of ablation, with different degrees of ablation occupying different positions in this space; it provides structured input data for training machine learning algorithms; and it establishes a mapping relationship from image features to the degree of ablation, which is the foundation for achieving automated evaluation.

[0064] The constructed ablation texture feature space is reduced in dimensionality and feature selected using principal component analysis (PCA) to generate an ablation feature spectrum characterizing the depth of ablation. PCA projects high-dimensional features into a low-dimensional space by identifying the main directions of change in the data, while retaining the main information of the original data. In cable ablation detection, PCA first calculates the covariance matrix of the feature space, then solves for eigenvalues ​​and eigenvectors, selects principal components based on the magnitude of the eigenvalues, and finally projects the original features into the principal component space. The dimensionality-reduced feature spectrum not only reduces the computational complexity of subsequent processing but also removes redundant information and noise interference, highlighting the key features of ablation depth more clearly.

[0065] The previously constructed multidimensional ablation texture feature space is standardized to eliminate the influence of different feature dimensions. Then, the covariance matrix of the feature space is calculated, reflecting the correlation and variation patterns among the features. Next, eigenvalue decomposition is used to solve for the eigenvalues ​​and eigenvectors of the covariance matrix. The magnitude of the eigenvalues ​​reflects the importance of the corresponding principal components, and the eigenvectors define the direction of the principal components. The principal components are selected based on the cumulative contribution rate of the eigenvalues, typically the top few principal components with a cumulative contribution rate of over 90%. Finally, the original features are projected onto the selected principal component space to obtain the dimensionality-reduced ablation feature spectrum. This feature spectrum has the following important characteristics: the dimensionality is significantly reduced while retaining the main information of the original features, typically from tens of dimensions to a dozen or so; redundancy and correlation between features are removed, and each principal component is mutually orthogonal; it highlights the key feature combinations that best reflect the differences in ablation depth; and it provides an optimized data format suitable for subsequent machine learning algorithms. The technical benefits are reflected in: significantly improving the computational efficiency of subsequent processing; enhancing the accuracy and stability of ablation degree assessment; providing comparable standardized feature representations for different ablation samples; and establishing an effective conversion mechanism from complex image features to concise feature spectra, which is a crucial link in the entire detection process.

[0066] For example, after the detection equipment acquires an image of a cable sample that has been ablated by a high-temperature flame, it decomposes the image into three frequency domain components using variational multi-scale decomposition. Analysis reveals that the energy distribution of the high-frequency component is concentrated in the ablation edge region, the mid-frequency component mainly covers the entire ablation area, and the low-frequency component contains a large amount of background information. Based on the ablation feature frequency domain, the system calculates a gradient map and determines a gradient threshold of 0.15 based on local region statistics, selecting 27 initial seed points to form a multi-core growth region. The tensor projection algorithm is applied to analyze the structural characteristics of the local region, constructing a principal direction vector field that clearly displays the directional information of the ablation edge. Using the vector field as a constraint to guide anisotropic diffusion filtering, it successfully suppresses false boundaries caused by uneven reflection on the cable surface, forming an accurate ablation region contour. Thirteen gray-level co-occurrence matrix features, 56 local binary mode features, and eight higher-order moment statistical features are extracted from the contour region, constituting a 77-dimensional original feature space. Principal component analysis was used to select the top 12 principal components as ablation characteristic spectra. The cumulative contribution rate of these principal components reached 94.8%, which effectively preserved the original characteristic information and significantly reduced the dimensionality, providing a reliable characteristic representation for subsequent ablation degree assessment.

[0067] In the specific implementation process, before training the image recognition model using the above training set to obtain the degree recognition model, the above method also includes the following steps: obtaining the image recognition model to be trained, wherein the image recognition model to be trained is a residual network model; adding an attention mechanism module between the convolutional layer and the activation function layer of the residual block of the image recognition model to be trained, so that the output of the convolutional layer is used as the input of the attention mechanism module, and the output of the attention mechanism module is used as the input of the activation function layer to obtain the image recognition model.

[0068] In this scheme, the residual network model maintains the training stability and performance of deep networks through residual connections. However, traditional residual network models treat all input features equally and lack the ability to focus on key features. Therefore, this scheme adds an attention mechanism module to each residual block, enabling the model to automatically learn and adjust the weights of different channels or features, allocating more computational resources to the regions where ablation features are most obvious, thereby improving the accuracy of feature recognition.

[0069] Current evaluation models are mostly static and cannot fully consider the temporal development characteristics of ablation. Specifically, based on the aforementioned ablation feature spectrum, an improved residual attention network is used to establish an ablation severity assessment model. This model is trained through transfer learning to form an ablation grading engine, including: spectral domain enhancement of the ablation feature spectrum; separation of ablation-specific frequency band information through multi-kernel orthogonal transformation; importing the ablation-specific frequency band information into a bidirectional gated feature selection unit to extract a discriminative feature set; constructing a multi-layer residual block structure for the discriminative feature set, introducing a channel attention mechanism in each residual block to form an adaptive feature expression; combining the adaptive feature expression with a pre-trained knowledge base for transfer weight mapping to generate an ablation knowledge prior model; optimizing the parameters of the ablation knowledge prior model through a cross-validation fine-tuning strategy to establish ablation category distinction boundaries; and combining the ablation category distinction boundaries with the ablation-specific frequency band information to construct an interpretable reasoning path, forming an ablation grading engine (i.e., a severity recognition model).

[0070] Specifically, spectral enhancement is performed on the ablation feature spectrum. Spectral enhancement refers to techniques for processing and optimizing signals in the frequency domain, aiming to strengthen the frequency components containing important information. In this process, multi-kernel orthogonal transform technology is applied to perform in-depth processing of the ablation feature spectrum. Multi-kernel orthogonal transform is a signal processing method that analyzes signals simultaneously through orthogonal transformations with multiple different parameters, adapting to different types of ablation features. This technique first maps the ablation feature spectrum to the frequency domain using methods such as Fourier transform, then applies different filter banks to extract key frequency bands, and finally combines multiple filtering results to form an enhanced output. This processing can effectively separate the frequency band information unique to ablation, including high-frequency information of the ablation edge, mid-frequency information of material changes, and low-frequency information of overall damage, laying the foundation for subsequent detailed analysis.

[0071] Spectral domain enhancement is a technique for processing and optimizing signals in the frequency domain. The obtained ablation feature spectrum is mapped to the frequency domain using a Fast Fourier Transform (FFT) to obtain the feature data in the frequency domain representation. Then, a multi-kernel orthogonal transform technique is applied in the frequency domain. This technique uses multiple orthogonal transform kernel functions with different parameters to simultaneously analyze and process the frequency domain data. Each kernel function can extract feature information within a specific frequency range, and multiple kernel functions working in parallel can cover the entire spectrum. Next, the system identifies and enhances key frequency components related to the ablation features while suppressing noise and irrelevant frequency components. Finally, the enhanced frequency domain data is converted back to the time domain through an inverse transform to obtain the optimized ablation feature representation. The technical value of this process lies in: highlighting the most discriminative frequency components in the ablation features, improving the feature's expressive power; effectively removing frequency domain noise, improving data quality; providing clearer input data for subsequent feature separation and selection; and enhancing the feature differences between different ablation degrees, facilitating accurate classification.

[0072] The multi-kernel orthogonal transform uses multiple different types of transform kernel functions, including Gaussian kernels, Laplace kernels, and polynomial kernels. Each kernel function has different response characteristics for signals in different frequency ranges. The system applies these kernel functions in parallel to transform the spectral domain-enhanced data, with each kernel function outputting a sub-band transform result. Then, energy analysis and correlation analysis are used to identify which sub-bands contain ablation-specific information: ablation-specific frequency bands typically have characteristics such as high energy density, specific frequency distribution patterns, and high correlation with known ablation samples. The system automatically selects sub-bands that meet these conditions and combines them to obtain ablation-specific frequency band information. At the same time, frequency bands that mainly contain background information, noise, or irrelevant material properties are filtered out. The key issues addressed by this separation technique include: eliminating the interference of inherent material properties on ablation detection, enabling the system to focus on changes caused by ablation; extracting the feature information that best reflects the degree and type of ablation, significantly improving the targeting of subsequent analysis; establishing a clear correspondence between ablation features and frequency domain characteristics, laying the foundation for automated identification; and significantly improving the distinguishability between different ablation samples, providing high-quality input data for accurate grading.

[0073] The separated ablation-specific frequency band information is imported into a bidirectional gated feature selection unit to extract a discriminative feature set. The bidirectional gated feature selection unit is a feature filtering mechanism that adaptively selects the most discriminative features. This unit contains two channels: a forward feature stream and a reverse feature stream. The forward stream extracts features progressively from lower to higher layers, while the reverse stream provides feedback guidance from higher to lower layers. The two channels interact via a gating mechanism, allowing the feature selection process to comprehensively consider information from different levels. In cable ablation detection, the bidirectional gated mechanism can accurately identify key feature differences at different ablation levels. For example, slight ablation mainly manifests as surface color changes, moderate ablation shows obvious texture changes, while severe ablation may lead to changes in material structure. Through this mechanism, the system can extract the most discriminative feature subset from the ablation-specific frequency band information.

[0074] The primary purpose of the bidirectional gating feature selection unit is indeed to find different types of data feature sets, but this "differentiation" has deeper technical implications and significant practical value. Specifically, this unit seeks feature subsets with different discriminative abilities and complementarities, rather than simple data classification. The working principle of the bidirectional gating mechanism is as follows: the forward channel starts from low-level features, gradually abstracting and combining them to extract composite features with high-level semantics; the reverse channel starts from high-level features, passing down guiding information to aid in the selection and optimization of low-level features. The gating mechanism establishes dynamic information exchange between the two channels, ensuring that the feature selection process considers both local details and global structure. The different feature sets sought in this process include: a detailed feature set, mainly containing local information such as the microscopic texture and edge details of the ablated surface; a structural feature set, containing geometric information such as the shape, distribution, and connectivity of the ablated region; an intensity feature set, reflecting the depth of ablation and the severity of material changes; and a temporal feature set, which, although a static image, can infer the development stage of ablation through multi-scale analysis. The technical significance of selecting these different feature sets is that different feature sets describe the ablation state from different perspectives, providing comprehensive and complementary information; enhancing the model's ability to understand complex ablation patterns; improving the robustness of classification decisions and avoiding errors that may occur when relying on a single feature type; and providing rich and structured input data for subsequent deep learning networks.

[0075] A multi-layer residual block structure is constructed from the extracted discriminative feature set, with a channel attention mechanism introduced in each residual block to form an adaptive feature representation. The residual block structure is a commonly used network building unit in deep learning, and by adding skip connections, it effectively solves the problem of training difficulties in deep networks. In the cable ablation detection system, the residual block is designed to contain a combination of convolutional layers, batch normalization layers, and activation functions, with a channel attention mechanism introduced in each residual block. The channel attention mechanism adaptively adjusts the contribution of different features by learning the importance weights of each feature channel. Specifically, the features are first subjected to global average pooling to obtain channel descriptors, then two fully connected layers are used to generate channel weights, and finally, the original features are weighted. This structure can automatically adjust the feature representation according to the characteristics of different ablation samples, enhancing the model's ability to perceive subtle differences in the degree of ablation.

[0076] The channel attention mechanism is added inside each residual block of the residual network, specifically after the convolutional layer and before the activation function, running in parallel with the residual connections. The detailed network structure is as follows: each residual block contains two main convolutional layers. The first layer extracts features, and the second layer integrates them. After the output of the second convolutional layer, the feature tensor is fed into two parallel processing paths: one path directly connects to the residual addition operation, and the other path passes through the channel attention module. The channel attention module performs global average pooling on the feature tensor, compressing the two-dimensional feature map of each channel into a scalar value, representing the global response strength of that channel. These scalar values ​​are then processed through two fully connected neural networks. The first layer uses the ReLU activation function for non-linear transformation, and the second layer uses the Sigmoid activation function to output weights between zero and one. These weights represent the importance of each channel; important channels have weights close to one, and unimportant channels have weights close to zero. Finally, these weights are multiplied element-wise with the original feature tensor at the channel level to achieve adaptive weighting of the features. The technical principle behind this design is that it enables the network to automatically learn which feature channels are most important for the current task and dynamically adjust the contribution of different channels. For ablation detection tasks, some channels may specifically respond to material color changes, some channels respond to texture changes, and some channels respond to shape features. The attention mechanism can automatically strengthen the most relevant channels according to the characteristics of specific samples. This adaptive feature representation significantly improves the network's ability to perceive subtle differences in ablation, enabling the model to more accurately distinguish different degrees of ablation.

[0077] This method combines adaptive feature representation with a pre-trained knowledge base for transfer weight mapping, generating a priori ablation knowledge model. The pre-trained knowledge base is a set of model parameters pre-trained on a large number of cable ablation samples, containing rich knowledge of the ablation characteristics of cable materials. Transfer weight mapping refers to selectively transferring the weight parameters of the pre-trained model to the current task model, accelerating model convergence and improving performance. In cable ablation detection, the network layers in the pre-trained knowledge base are first analyzed to identify key layers related to ablation characteristics; then, a weight mapping strategy is designed to map the weight parameters of these key layers to the corresponding positions in the current model; finally, a parameter fusion algorithm is used to integrate the pre-trained knowledge with the characteristics of the current task, forming an ablation model with domain prior knowledge. This method effectively solves the problem of insufficient cable ablation samples, enabling the model to quickly adapt to new ablation types based on existing knowledge.

[0078] The above describes the transfer weight mapping process, the core principle of which is to transfer the general feature extraction capabilities pre-trained on a large-scale dataset to a specific ablation detection task. The specific technical principle is as follows: the pre-training knowledge base contains neural network models trained on a large number of cable ablation samples. These models have learned to recognize various ablation patterns and features. The transfer weight mapping process analyzes the network structure of the pre-trained models, identifying which layers are mainly responsible for low-level feature extraction (such as edges and textures) and which layers are responsible for high-level feature combination (such as shape and pattern). Then, based on the characteristics of the current task, the weight parameters of relevant layers in the pre-trained models are selectively copied. For low-level feature extraction layers, weights are usually directly copied because the basic features learned by these layers (such as edge detectors and texture filters) have good generality. For high-level feature layers, adaptive adjustments are needed, using parameter fusion, weight scaling, and other methods to adapt to new task requirements. The specific implementation includes: establishing the layer correspondence between the current network structure and the pre-trained model; then designing a weight mapping strategy to determine which weights are directly copied, which need adjustment, and which need to be reinitialized; next, performing parameter passing to copy the selected weight parameters to their corresponding positions in the current model; and finally, fine-tuning and optimization to precisely adjust the passed-in weights on new data. The value of this technique lies in: significantly reducing the amount of data and computation time required for model training; inheriting the rich knowledge and mature feature representation capabilities of the pre-trained model; improving the model's adaptability and generalization ability to new ablation types; and solving the problem of insufficient training samples in specific application scenarios.

[0079] The ablation knowledge prior model is optimized using a cross-validation fine-tuning strategy to establish ablation category distinction boundaries. Cross-validation fine-tuning is a model parameter optimization technique that evaluates model performance through multiple validations with different data partitions, avoiding overfitting. In the cable ablation detection system, a k-fold cross-validation method is used, dividing the available ablation sample dataset into k subsets. Each time, k-1 subsets are used for training, and the remaining subset is used for validation. For each fold validation, the system uses gradient descent with an early stopping strategy to adjust the model parameters, avoiding overfitting to a single training set. By synthesizing the multi-fold validation results, the system determines the optimal model parameter configuration and retrains the complete model using these parameters, establishing accurate ablation category distinction boundaries. These boundaries can accurately distinguish between different levels of ablation, such as minor, moderate, and severe ablation, providing a basis for ablation risk assessment.

[0080] The ablation category distinction is achieved based on decision boundaries in the feature space, a process closely and complexly related to the previously extracted ablation boundary information. Specifically, the system establishes multiple decision hyperplanes in the high-dimensional feature space, each dividing the feature space into different ablation category regions. These decision boundaries are determined through training machine learning algorithms (such as support vector machines and neural network classifiers) on a large number of labeled samples. During training, the algorithm learns the distribution patterns of samples with different ablation degrees in the feature space, finding the optimal classification boundary. The position and shape of the classification boundary reflect the degree of difference in feature representation between different ablation categories. This category distinction is indeed significantly related to the physical ablation boundary: the physical ablation boundary provides spatial location information, telling the system which areas have been ablated; while the category distinction boundary in the feature space provides severity information, telling the system the severity of these ablation areas. Using both together enables a complete assessment that knows both "where ablation occurs" and "how severe the ablation is." The subsequent implementation includes: projecting the feature vector of the new ablation sample into the established feature space; determining the category region it belongs to based on the position of the feature vector in the space; combining physical boundary information to determine the specific ablation region and the corresponding ablation level; and outputting a complete ablation assessment result containing both location and degree information. The technical problems solved by this method include: realizing the transformation from qualitative judgment to quantitative assessment; establishing a mathematical mapping relationship between image features and ablation degree; and providing a repeatable and quantifiable ablation assessment standard.

[0081] By combining the ablation category distinction boundaries with ablation-specific frequency band information, an interpretable reasoning path is constructed to form an ablation grading engine. An interpretable reasoning path refers to a technical mechanism that can track and explain the model's decision-making process, making the model's classification results understandable. In the cable ablation detection system, the ablation category distinction boundaries are first mapped back to the feature space, establishing a correspondence between features and ablation levels. Then, key features from the ablation-specific frequency band information are combined to construct a feature activation mapping, visualizing the discrimination regions for different ablation levels. Finally, interpretable models such as decision trees are used to assist in analyzing the decision-making basis of the deep model, forming a complete reasoning and explanation chain. This method enables the system not only to output the classification results of the ablation degree but also to explain the basis for this classification, such as indicating that changes in material structure or color anomalies in a certain area are key factors in judging the ablation degree. The resulting ablation grading engine can accurately, reliably, and interpretably assess the ablation degree of the cable's water-blocking buffer layer.

[0082] The system constructs a feature-to-decision mapping path, recording each intermediate step from the original ablation feature input to the final classification output, including feature transformation, weight calculation, and activation function response. Each processing step retains detailed mathematical operation records and parameter change trajectories. Next, it analyzes the importance of key features, calculating the contribution of each input feature to the final decision and identifying the key features with the greatest impact on the classification result. The importance weights of different features are quantified through gradient analysis and feature ablation experiments. Then, the system visualizes the decision logic, transforming the complex mathematical calculation process into intuitive graphical representations such as decision trees, heatmaps, and feature activation maps. These visualization tools clearly demonstrate the key information upon which the model makes its judgments. The ablation grading engine is a comprehensive process: it integrates the technical achievements of all preceding steps, including feature extraction, classification decision, and interpretable analysis; it provides a processing pipeline from input to output, automatically completing the entire process from ablation image to grading result; and it also possesses interpretation and verification functions, allowing users to view the detailed basis and reasoning process for any classification decision. The key technical issues addressed by this hierarchical engine include: providing transparent and reliable automated ablation assessment capabilities; establishing a human-computer interaction verification mechanism, allowing experts to check and confirm the model's judgment criteria; enhancing the system's maintainability and improveability, enabling the identification of model deficiencies and targeted optimization through analysis of the reasoning path; and providing technical support for the standardization and normalization of ablation detection, resulting in a repeatable and verifiable assessment process.

[0083] For example, when the system processes a set of pre-processed cable ablation sample data, it first decomposes the ablation feature spectrum into three sub-bands using a multi-kernel orthogonal transformation: the low-frequency band reflects the overall ablation area, the mid-frequency band reflects the ablation depth variation, and the high-frequency band reflects surface microstructure damage. After processing by a bidirectional gating unit, these frequency bands extract 32 highly discriminative features. These features are then processed through a four-layer residual block structure, where the channel attention mechanism in each residual block automatically adjusts the weights of each feature based on the sample characteristics, enhancing the ability to identify key ablation features. The system selects model parameters trained on over 10,000 cable ablation samples from a pre-training knowledge base and integrates them into the current network through transfer mapping, significantly improving the model's adaptability to new ablation types. Five-fold cross-validation is used to optimize the model, determining an optimal learning rate of 0.001 and an optimal regularization coefficient of 0.0005, successfully establishing a classification boundary capable of distinguishing between mild, moderate, and severe ablation states. The resulting ablation classification engine not only achieves an accuracy of 92.7%, but also visually displays key areas that influence classification decisions through feature activation mapping, providing cable maintenance personnel with a reliable basis for decision-making.

[0084] In some embodiments, the visible light image and the infrared image are weighted and fused to obtain a target image. This can be achieved through the following steps: preprocessing the visible light image to obtain a preprocessed visible light image, wherein the preprocessing method includes filtering and / or image enhancement; preprocessing the infrared image to obtain a preprocessed infrared image, wherein the preprocessing method includes filtering and / or image enhancement; obtaining a first weight coefficient and a second weight coefficient, wherein the first weight coefficient is the weight coefficient of the preprocessed visible light image, and the second weight coefficient is the weight coefficient of the preprocessed infrared image; and weighting and fusing the preprocessed visible light image and the preprocessed infrared image according to the first weight coefficient and the second weight coefficient to obtain the target image.

[0085] In this scheme, the preprocessing step removes noise from the image using filtering algorithms, improving the signal-to-noise ratio and making the features of the ablation area more prominent. Even with slight texture variations on the cable surface or inconsistent lighting conditions, the ablation area can be accurately detected. Image enhancement techniques, such as contrast stretching and histogram equalization, further enhance the contrast and clarity of the ablation area, making it more visible in the image and facilitating subsequent feature extraction and deep learning model recognition. The weighted fusion process dynamically adjusts the weights of the visible light and infrared images, ensuring that the ablation features are fully represented in the target image under different detection conditions. This avoids the limitations of information from a single image source and improves the comprehensiveness and accuracy of ablation degree detection.

[0086] Specifically, a dual-spectral imaging device is used to simultaneously acquire visible and infrared band data of the cable's water-blocking buffer layer. After nonlinear contrast optimization processing, an enhanced bispectral image is obtained, including: simultaneously acquiring cable surface images in the visible and mid-infrared bands using a dual-sensor array to form raw dual-channel data; using a polarization filtering matrix to eliminate surface reflection interference and extracting ablation-specific spectral response signals from the raw dual-channel data; performing multi-scale decomposition of the spectral response signals using wavelet transform to separate thermal and mechanical damage information layers; applying a local frequency domain enhancement algorithm to the thermal and mechanical damage information layers to highlight ablation boundary features; and integrating the spectral response signals and ablation boundary features using an adaptive weighted fusion processor to generate an enhanced bispectral image with the ability to distinguish between shallow and deep layers.

[0087] Specifically, images of the cable surface in the visible and mid-infrared bands are acquired simultaneously using a dual-sensor array. The dual-sensor array consists of a precisely calibrated and aligned visible light camera (operating wavelength 400-700nm) and a mid-infrared camera (operating wavelength 3-5μm). The two sensors are optically coaxial to ensure complete overlap of the image fields of view. This design enables the system to simultaneously capture the color and texture features (visible light) and thermal radiation distribution features (mid-infrared) of the cable surface, forming a raw dual-channel data matrix.

[0088] The dual-sensor array system comprises two independent imaging sensors. The visible light camera is equipped with a CCD or CMOS sensor, operating in a strictly limited wavelength range of 400-700nm, with a pixel resolution typically of 1920×1080 or higher. Each pixel records the grayscale value of the RGB three channels in 8-bit or 16-bit integer format. The mid-infrared camera uses an InGaAs or MCT detector, operating in a precisely controlled wavelength range of 3-5μm, with the same pixel resolution as the visible light camera. Each pixel records the radiation intensity value corresponding to the temperature in floating-point format. The two sensors achieve strict optical axis alignment through a precise optical beam splitter system, ensuring that the pixel coordinates of the same physical point correspond perfectly on both sensors. During data acquisition, the system employs a hardware-triggered synchronization mechanism to ensure that both sensors are exposed at the same time, achieving microsecond-level time synchronization accuracy. The acquired raw data consists of two independent image matrices: a visible light image matrix V(i,j,c), where i and j are spatial coordinates and c is the RGB color channel; and an infrared image matrix IR(i,j), recording the thermal radiation intensity of each pixel. After spatial registration and temporal alignment, these two matrices are merged to obtain a four-dimensional data tensor D(i,j,k), where k=1,2,3 correspond to the RGB channels and k=4 corresponds to the infrared channel. This is the original dual-channel data.

[0089] A polarization filter matrix is ​​applied to the original dual-channel data to eliminate surface reflection interference. The polarization filter matrix consists of linear polarization filters at multiple angles (typically 0°, 45°, 90°, and 135°). By measuring the intensity of reflected light in different polarization directions, Stokes parameters are calculated, thereby separating the specular reflection component from the material's own scattering component. This step effectively removes the gloss and reflection from the cable surface, making the ablation-specific spectral response signal more prominent.

[0090] The interference to be eliminated here is specular reflection from the cable surface. The polarization filter matrix system contains four linear polarization filters, installed in the optical path at angles of 0°, 45°, 90°, and 135°, respectively. Each filter can be independently rotated into the optical path. In actual processing, the system sequentially uses these four polarization angles to image the same area, obtaining four polarization images I0, I...45 I 90 I 135 Then, the data is processed using the Stokes parameter calculation formula: I = I0 + I 90 (Total light intensity), Q = I0 - I 90 (Horizontal-vertical polarization difference), U=I 45 -I 135 (±45° polarization difference), V=0 (circular polarization component, zero under linear polarization). Next, calculate the degree of polarization DOP=√(Q 2 +U 2 +V 2 The polarization angle AOP is 0.5 × arctan(U / Q). Specular reflection has a high degree of polarization (DOP > 0.7), while material scattering has a lower degree of polarization (DOP < 0.3). By setting a threshold separation algorithm, for each pixel, if DOP > 0.5, the pixel is determined to be dominated by specular reflection, and the system calculates the specular reflection component according to Fresnel's law of reflection and subtracts it from the total intensity; if DOP < 0.5, the original intensity value is retained. In the final output spectral response signal matrix S(i,j,λ), the value of each pixel is the pure material spectral characteristic after removing specular reflection. The first sentence is the original multispectral image data matrix including specular reflection, and the second sentence is the pure material spectral response vector after polarization analysis and reflection separation.

[0091] The extracted spectral response signal is decomposed into multiple scales using wavelet transform. Wavelet transform employs wavelet basis functions at different scales (such as Daubechies or Haar wavelets) to perform joint time-frequency analysis on the signal, decomposing the image into low-frequency approximate components and multiple high-frequency detail components at different scales. By analyzing the energy distribution and correlation of different frequency bands, the system can distinguish between thermal damage (typically manifested as characteristic hotspots in the mid-infrared band) and mechanical damage (typically manifested as structural changes in the visible light band), forming separate information layers. A local frequency domain enhancement algorithm is applied to the separated thermal and mechanical damage information layers to highlight the ablation boundary features. The local frequency domain enhancement algorithm selectively enhances specific frequency components based on the spectral characteristics of local regions using an adaptive frequency domain filter. For ablation boundaries, the algorithm focuses on enhancing the mid-to-high frequency components, making the boundaries sharper and more distinct. Specifically, the algorithm adaptively adjusts the detail coefficients in the wavelet domain, enhancing coefficients representing boundaries and suppressing coefficients representing noise.

[0092] The separation of thermal and mechanical damage is achieved through multi-scale decomposition using wavelet transform. The specific data processing algorithm is as follows: The system performs Discrete Wavelet Transform (DWT) on the spectral response signal S(i,j,λ) after removing reflection interference, selecting the Daubechies db4 wavelet as the basis function due to its good time-frequency localization characteristics. The wavelet decomposition process is: W(a,b)=∫S(x)ψ*(a,b)(x)dx, where S(x) is the original signal or image, ψ*(a,b) is the wavelet basis function, a is the scale parameter, and b is the displacement parameter. The system performs 4 levels of wavelet decomposition, obtaining one low-frequency approximation coefficient matrix A4 and four high-frequency detail coefficient matrices D1, D2, D3, and D4. Next, frequency domain energy analysis is performed: the energy distribution of each decomposition layer is calculated as Ek=∑∑|Wk(i,j)|2, where k represents different decomposition layers. The damage is classified based on the differences in physical properties between thermal and mechanical damage: Thermal damage mainly manifests as changes in the uniformity of the material, such as softening and carbonization. In the frequency domain, this type of damage is mainly characterized by low- and mid-frequency components (corresponding to layers A4, D4, and D3), with high energy concentration and continuous distribution. Mechanical damage, such as cracks, scratches, and indentations, has sharp boundary features and is mainly characterized by high-frequency components (corresponding to layers D1 and D2) in the frequency domain, with pulsed energy distribution. The system automatically separates these damages using an energy threshold determination algorithm: An energy ratio threshold Rth = 0.6 is set. If the low-frequency energy ratio Rlow = (EA4 + ED4 + ED3) / (Etotal) > Rth, where Etotal is the total energy, then the region is determined to be dominated by thermal damage; if the high-frequency energy ratio Rhigh = (ED1 + ED2) / (Etotal) > Rth, then it is determined to be dominated by mechanical damage. Furthermore, the system verifies the separation results through correlation analysis: The cross-correlation function Rxy(τ) = ∫x(t)y(t+τ)dt is calculated between different frequency bands, where x(t) represents the signal of a specific ablation feature (such as the degree of thermal damage, mechanical damage distribution, etc.) changing with time or location at the first time point, t represents time, y(t) represents the signal of a specific ablation feature (such as the degree of thermal damage, mechanical damage distribution, etc.) changing with time or location at the second time point, and τ represents the time delay parameter. The low-frequency components in the thermal damage region show high correlation (R>0.8), while the high-frequency components in the mechanical damage region show low correlation but significant peaks (R<0.5 but with significant peaks). Finally, two independent feature layers are output: the thermal damage information layer contains reconstructed low-to-mid-frequency components, and the mechanical damage information layer contains reconstructed high-frequency components.

[0093] The algorithm operates in the wavelet transform domain, processing the previously separated thermal and mechanical damage information layers separately. For each wavelet coefficient W(i,j,k), the system calculates its local statistical characteristics within a 3×3 neighborhood window: the local mean μlocal=(1 / 9)∑∑W(i+m,j+n,k), where i, m, j, n, and k are constants, local represents the neighborhood window, W refers to the image's grayscale value or other data feature value, and i+m,j+n, and k are the spatial or frequency domain offset operations performed on a point (i,j) in the image, forming a neighborhood centered on that point. m and n are the offsets in the row and column directions, respectively, used to traverse all point positions within the neighborhood. The local variance σ 2 local=(1 / 9)∑∑[W(i+m,j+n,k)-μlocal] 2 Local gradient magnitude Next, we calculate the adaptive enhancement coefficient: α(i,j)=1+β×exp(-σ 2 local / σ 2 threshold)×(Glocal / Gmax), where β is the reinforcement strength parameter (usually taken as 1.5-2.5), σ 2 `threshold` is the variance threshold, `Glocal / ` is the sum of gradient magnitudes, and `Gmax` is the global maximum gradient value. For boundary regions (high gradient, low variance), the value of α is close to the maximum value; for smooth regions (low gradient, high variance), the value of α is close to 1. The enhancement formula is: `Wenhanced(i,j,k) = α(i,j) × W(i,j,k)`. The specific extraction process of the ablation boundary is more complex: the enhanced signal is reconstructed through inverse wavelet transform: `Senhanced = IDWT(Wenhanced)`, where `Senhanced` represents the enhanced signal or image, `IDWT` represents the inverse discrete wavelet transform, and `Wenhanced` represents the coefficients enhanced in the wavelet domain; then, the Laplacian edge detection operator is applied. The second derivative is calculated; then, an adaptive threshold Tadaptive = μglobal + k × σglobal (k is usually 2-3) is used for edge binarization, where μglobal represents the global mean and σglobal represents the global standard deviation; finally, morphological operations (opening and closing operations) are used to remove noise and connect broken boundaries. After the boundary feature extraction is completed, the system fuses the enhanced spectral response signal with the boundary features. The fusion weight is calculated based on the local information entropy H = -∑pi × log2(pi) and the structural similarity index SSIM, where pi represents the probability.

[0094] The spectral response signal and ablation boundary features are integrated using an adaptive weighted fusion processor. Based on local information entropy and gradient magnitude, the adaptive weighted fusion processor calculates the optimal fusion weight for each pixel location, weighting the enhanced boundary features with the original spectral response signal. The fusion process considers information complementarity and redundancy, ensuring that while preserving complete spectral information, it enhances the boundary and depth-level features of the ablation region, ultimately generating an enhanced bispectral image with depth-level discrimination capabilities, laying the foundation for subsequent precise ablation region localization and feature extraction.

[0095] Multiple weighting factors are calculated for each pixel position (i,j). The local entropy is calculated using a sliding window method: within a 5×5 window centered at (i,j), the probability distribution p(g) of the grayscale values ​​is statistically analyzed, and then the entropy H(i,j) is calculated as -∑(g=0 to 255)p(g)×log2(p(g)). The gradient magnitude is calculated using the Sobel operator: Gx=Sobelx*I(i,j), Gy=Sobelly*I(i,j), G(i,j)=√(Gx) / I(i,j ... 2 +Gy 2The weighting factors are normalized as follows: Hnorm(i,j) = (H(i,j) - Hmin) / (Hmax - Hmin), Gnorm(i,j) = (G(i,j) - Gmin) / (Gmax - Gmin). The overall weight is calculated using a weighted combination: w(i,j) = α × Hnorm(i,j) + β × Gnorm(i,j) + γ × C(i,j), where α, β, and γ are weighting coefficients (α = 0.4, β = 0.3, γ = 0.3), and C(i,j) is the local contrast factor: C(i,j) = (Imax - Imin) / (Imax + Imin). The final pixel-level fusion formula is: F(i,j)=w(i,j)×Sboundary(i,j)+[1-w(i,j)]×Sspectral(i,j)+ε×N(i,j), where Sboundary is the ablation boundary feature value, Sspectral is the spectral response signal value, ε is the noise suppression coefficient (usually 0.05-0.1), and N(i,j) is the local noise estimate. This fusion process considers the complementarity of information: in the boundary region, the boundary feature has a higher weight to maintain edge sharpening; in the uniform region, the spectral signal has a higher weight to maintain the integrity of spectral information. At the same time, redundancy analysis is used to avoid information duplication: the mutual information I(Sboundary; Sspectral) of the two signals is calculated as I(Sboundary; Sspectral)=∑∑p(x,y)×log[p(x,y) / (p(x)×p(y))], and the weight allocation is automatically adjusted when the mutual information is too high. The resulting enhanced bispectral image matrix has the following characteristics: it retains the physical meaning of the original spectrum, enhances the boundary clarity of the ablation region, and has the ability to distinguish between light and dark layers.

[0096] In the specific implementation process, after determining the degree of ablation of the water-blocking buffer layer of the cable, the method further includes the following steps: using Fourier transform to convert the target image from the time domain to the frequency domain to obtain a frequency domain target image; using an unwrapping algorithm to process the frequency domain target image to obtain a depth frequency domain target image; using an edge enhancement algorithm to process the frequency domain target image to obtain a micro-frequency domain target image; performing weighted fusion of the depth frequency domain target image and the micro-frequency domain target image to obtain an ablation image, and displaying the ablation image on the display interface.

[0097] In this scheme, Fourier transform extracts frequency component information from the target image, providing a foundation for subsequent depth information extraction and edge enhancement. The unwrapping algorithm processes the phase information obtained from the Fourier transform, generating a continuous depth map, avoiding discontinuities in depth information, and improving the accuracy of depth assessment. The edge enhancement algorithm enhances the edge information in the image, making the ablation edges and microstructures of the cable surface clearer, facilitating subsequent ablation feature spectrum extraction. By weighted fusion of the depth-frequency domain target image and the micro-frequency domain target image, a high-precision ablation image can be generated, which clearly displays the depth information and edge features of the ablation area.

[0098] Current solutions for 3D morphology reconstruction technology have insufficient application in cable ablation detection, making it difficult to accurately quantify microscopic morphological features. Specifically, based on the output results of the ablation grading engine and combined with phase structured light coding technology, a high-precision ablation depth map is constructed, including: using the output results of the ablation grading engine as region guiding constraints to mark key ablation detection intervals; projecting a composite sinusoidal phase-coded grating onto the key ablation detection intervals to generate a variable-density stripe projection pattern; capturing multi-angle reflection responses from the variable-density stripe projection pattern and extracting phase information through Fourier transform; processing the phase information using a Gaussian-Helmholtz phase unwrapping algorithm to eliminate phase jumps and construct a continuous depth map; spatially registering the continuous depth map with the key ablation detection intervals and highlighting microscopic textures using a Laplacian edge enhancement algorithm; fusing the microscopic textures with the continuous depth map and using anisotropic filtering to preserve structural boundaries to generate a high-precision ablation depth map (i.e., an ablation image).

[0099] Specifically, the output of the ablation grading engine is used as a region-guided constraint to mark key ablation detection intervals. Region-guided constraints utilize the classification results from the previous stage of the ablation grading engine to determine areas requiring focused attention, thereby improving the accuracy and efficiency of subsequent 3D reconstruction. Specifically, the ablation grading engine output includes ablation level information and corresponding confidence scores for each region on the cable surface. By setting a confidence threshold, high-confidence ablation regions are extracted as key detection intervals. These intervals typically include three parts: the ablation boundary transition zone, the ablation core zone, and the potential development zone, each with different detection importance and characteristic features. The marking process employs a boundary expansion algorithm, setting a buffer zone of a certain width around the determined ablation region to ensure complete capture of the transitional features of ablation evolution.

[0100] The three zones of the key ablation detection area are defined based on the output of the ablation grading engine and the physical characteristics of cable ablation. Specifically, they include: the ablation boundary transition zone, the ablation core zone, and the potential development zone. The ablation boundary transition zone refers to the boundary area where normal material transitions to ablated material. This area is characterized by a large gradient in material properties and is a key location for judging the degree and trend of ablation. The ablation core zone is the most severely ablated central area, where significant physicochemical changes have occurred. It is the main reference area for assessing the current damage level. The potential development zone refers to the area around the ablation core zone where ablation may continue to develop. Although the current changes are not obvious, it has a high risk of ablation propagation. The three intervals are obtained as follows: First, preliminary region division is performed based on the classification results and confidence scores output by the ablation grading engine. Then, high-confidence ablation areas are screened by setting a confidence threshold, typically selecting areas with a confidence score higher than 0.85 as key detection targets. Next, a boundary expansion algorithm is applied to set a buffer zone of a certain width around the determined ablation area to ensure complete capture of the transitional features of ablation evolution. Finally, based on the physical development law of ablation, the detection area is subdivided into the three functional sub-intervals mentioned above. This interval division solves the following technical problems: it avoids indiscriminate high-precision detection of the entire cable surface, significantly improving detection efficiency; it ensures the detection accuracy of key areas, focusing on the parts that require the most monitoring.

[0101] A composite sinusoidal phase-coded grating is projected onto the marked critical ablation detection area to generate a variable-density fringe projection pattern. The composite sinusoidal phase-coded grating is a structured light projection technique that acquires three-dimensional information of the target surface by projecting grating fringes with specific coding rules. In cable ablation detection, this technique uses a 632.8nm laser source to generate a base grating, which is then modulated into a composite phase-coded mode by a digital optical processor. Composite phase coding refers to the simultaneous use of multiple sets of sinusoidal gratings with different frequencies and phases for encoding. This composite coding method can generate a variable-density fringe projection pattern, automatically adjusting the fringe density in areas with different ablation severity. High-density fringes are used in fine ablation areas to improve depth resolution, while low-density fringes are used in rough ablation areas to ensure correct fringe identification.

[0102] The above paragraph describes the projection technique of a composite sinusoidal phase-coded grating, a crucial component of structured light 3D reconstruction technology. Specifically, the composite sinusoidal phase-coded grating is a specially designed optical projection pattern that simultaneously contains multiple sinusoidal fringes of different frequencies and phases. This composite coding design aims to obtain richer depth information in a single projection, improving the accuracy and efficiency of 3D reconstruction. The projection process is implemented as follows: the system uses a digital light processor to generate a composite phase-coded pattern containing three sets of sinusoidal fringes with frequencies of 0.5 mm and 0.5 mm respectively. -1 1.0mm -1 and 2.0mm -1 Each set of stripes also contains four different phase steps with a phase difference of π / 2, forming a complete phase coding sequence. The projection system sequentially projects these coding patterns onto the surface of the critical ablation detection area. The formation mechanism of the variable density stripe projection pattern is as follows: the system automatically adjusts the stripe density according to different ablation severity, using high-density stripes in fine ablation areas to improve depth resolution, and using low-density stripes in rough ablation areas to ensure that the stripes can be correctly identified. This adaptive density adjustment ensures reliable phase information can be obtained under various ablation surface conditions. The technical value is reflected in: significantly improving the accuracy of 3D reconstruction, especially the reconstruction capability for complex surface morphologies; reducing the number of projections and improving detection efficiency; enhancing adaptability to surfaces with different ablation degrees; and providing high-quality raw data for subsequent depth information extraction.

[0103] Multi-angle reflection responses are captured from variable-density fringe projection patterns, and phase information is extracted using Fourier transform. Multi-angle reflection response capture refers to simultaneously acquiring fringe images of cable surface reflections using multiple calibrated cameras or a single camera at multiple positions sequentially. These images contain information about fringe deformation caused by variations in surface depth. The acquired image sequence is first preprocessed, including noise reduction, contrast enhancement, and geometric correction, and then phase analysis is performed using Fourier transform. Fourier transform is a mathematical tool for converting spatial domain signals to the frequency domain, and it is particularly effective for images containing periodic fringes. In cable ablation detection, two-dimensional Fourier transform converts the fringe image from the spatial domain to the frequency domain, displaying the frequency peaks corresponding to the fringes. By selecting the spectral component corresponding to the projection grating frequency through bandpass filtering and then performing an inverse Fourier transform, the complex signal containing phase information can be extracted. The phase information is obtained by calculating the argument of the complex signal.

[0104] The above paragraph describes the multi-angle reflection response capture and Fourier transform phase extraction techniques. The specific process of multi-angle reflection response capture is as follows: the system is configured with multiple precisely calibrated cameras, located at different spatial positions, which can simultaneously observe the stripe pattern projected on the cable surface from multiple perspectives; these cameras can be a fixed multi-camera array or a single camera with multi-position sequence acquisition; each camera at each angle will capture a stripe image modulated by surface depth changes, and the degree of stripe deformation directly reflects the three-dimensional morphology information of the surface. The technical principle of Fourier transform for extracting phase information is as follows: A two-dimensional fast Fourier transform is performed on the captured fringe image to convert the fringe pattern in the spatial domain to the frequency domain. In the frequency domain, the fringes correspond to specific frequency peaks. The system selects the spectral component corresponding to the frequency of the projected grating through bandpass filtering. Then, an inverse Fourier transform is performed on the selected spectral component to obtain a complex signal containing phase information. The phase information is obtained by calculating the argument of the complex signal, using the formula φ = arctan(imaginary part / real part). The fusion processing of multi-angle information involves synthesizing phase information obtained from different angles through a weighted average, with the weights dynamically determined based on the signal-to-noise ratio and reliability of each angle. This multi-angle fusion effectively reduces interference from factors such as shadows and reflections on phase extraction, improving the accuracy and completeness of phase information. Technical advantages include: significantly improved reliability of phase information, avoiding occlusion and reflection problems that may occur with single-angle acquisition; enhanced adaptability to complex surface geometries; provision of high-quality input data for subsequent phase unwrapping and depth reconstruction; and achievement of millimeter-level or even sub-millimeter-level depth measurement accuracy.

[0105] The phase information is processed using the Gauss-Helmholtz phase unwrapping algorithm to eliminate phase jumps and construct a continuous depth map. Phase unwrapping is a key step in structured light 3D reconstruction, used to solve the phase jump problem caused by the phase values ​​extracted by Fourier transform being restricted to the interval [-pi, pi]. The Gauss-Helmholtz phase unwrapping algorithm is an unwrapping method based on partial differential equations, obtaining a continuous phase field by solving the Helmholtz equations:

[0106]

[0107] Where Ψ(x,y) is the continuous phase field to be solved, and φ w (x,y) is the wrapping phase field. Let λ represent the Laplacian operator, where λ is the regularization parameter controlling the smoothness of the solution. The algorithm first calculates the gradient field of the wrapping phase, then applies a Gaussian filter to the gradient field to reduce noise, and finally obtains the continuous phase field by iteratively solving the aforementioned partial differential equation. The relationship between the continuous phase field and the depth value is calculated using a transformation function obtained from system calibration.

[0108]

[0109] Where D(x,y) is the surface depth value, b is the baseline distance between the projector and the camera, g is the camera focal length, and P is the spatial period of the projection grating. This transformation yields a continuous depth map of the cable ablation surface.

[0110] The above describes the phase jump elimination and continuous depth mapping construction techniques of the Gauss-Helmholtz phase unwrapping algorithm. The phase jump problem arises because the phase values ​​extracted by the Fourier transform are limited to the range of -π to π. When the actual phase exceeds this range, a periodic jump of 2π occurs, leading to discontinuities in depth information. This jump is particularly noticeable in ablation detection because ablated surfaces often exhibit significant depth variations. The specific working principle of the Gauss-Helmholtz algorithm is as follows: the phase unwrapping problem is transformed into a boundary value problem of solving the Poisson equation. The algorithm is based on the physical assumption that the true continuous phase field should be smoothly changing, and its Laplace operator is equal to the divergence of the wrapped phase gradient. The solution process uses the finite difference method to discretize the partial differential equations, and then solves the linear equation system through an iterative algorithm. The algorithm also introduces a regularization term to control the smoothness of the solution, balancing data fitting and solution continuity. The construction of the continuous depth mapping is achieved through the phase-depth transformation relationship. Technical features include: effectively solving the instability problem of traditional path tracing methods in noisy environments; having good noise resistance and being able to obtain reliable unwrapping results under complex surface conditions; and providing globally consistent phase unwrapping to avoid the propagation of local errors.

[0111] Spatial registration is performed between the continuous depth map and the key ablation detection region, and the Laplacian edge enhancement algorithm is used to highlight the micro-texture. Spatial registration is the process of accurately corresponding the 3D depth map with the 2D ablation region information, ensuring spatial consistency between depth information and ablation classification results. The registration process first establishes the transformation relationship between the 2D image coordinate system and the 3D spatial coordinate system, and then uses the iterative nearest point algorithm for fine alignment, ultimately achieving an accurate correspondence between the depth map and the key ablation detection region. After registration, the Laplacian edge enhancement algorithm is applied to highlight the micro-texture features of the ablated surface. The Laplacian operator is a commonly used second-order differential operator that can detect areas with abrupt changes in intensity in an image. In cable ablation detection, the Laplacian operator processes the depth map to highlight areas with large surface depth changes, thereby enhancing the visibility of micro-texture features.

[0112] The spatial registration and Laplacian edge enhancement described above are key technical steps for accurately mapping 3D depth information to 2D ablation region information. The specific process of spatial registration includes: establishing a mathematical transformation relationship between the 3D depth mapping coordinate system and the 2D ablation detection interval coordinate system; this transformation relationship includes geometric transformation parameters such as rotation, translation, and scaling, which are predetermined through a calibration process; then, a fine alignment is performed using an iterative nearest-point algorithm, which optimizes the transformation parameters by minimizing the distance between corresponding points; the registration process also considers geometric distortions caused by factors such as projection angle and camera viewpoint, compensating for these distortions through lens correction and perspective transformation; ultimately, precise spatial consistency between depth information and ablation classification results is achieved. The working mechanism of the Laplacian edge enhancement algorithm is as follows: the Laplacian operator is a second-order differential operator, and its calculation formula is... This operator can detect regions with abrupt changes in intensity in images, and is particularly sensitive to edge and detail features. Applying the Laplacian operator to the ablation depth map highlights areas with significant surface depth variations, which typically correspond to surface damage caused by ablation. The enhanced micro-texture information includes details such as surface roughness changes caused by ablation, microcracks and pores, and steps formed by material detachment. These micro-features are of significant value in assessing the severity and development trend of ablation. The technical significance lies in: achieving accurate fusion of multimodal data, ensuring the spatial correspondence between 3D morphology information and 2D classification results; significantly enhancing the visibility of micro-features on ablated surfaces and improving the detection capability of detail features; providing rich and accurate morphology data for subsequent comprehensive risk assessment; and establishing a complete detection chain from macro-classification to micro-morphology, realizing multi-scale ablation state characterization.

[0113] By fusing micro-textures with continuous depth maps and employing anisotropic filtering to preserve structural boundaries, a high-precision ablation depth map is generated. The fusion of micro-textures and depth maps involves comprehensively considering the results of Laplacian edge enhancement with the original depth map, aiming to highlight surface details while maintaining overall depth information. A weighted averaging method is used, with weights dynamically adjusted based on regional characteristics to ensure more detail is retained in edges and texture-rich areas. The fused result is further optimized using anisotropic filtering, an edge-preserving smoothing technique that smooths along structural boundaries while maintaining edge sharpness perpendicular to the boundaries. During anisotropic filtering, the diffusion coefficient decreases as the gradient magnitude increases, ensuring reduced diffusion at the boundaries. This process generates a high-precision ablation depth map that possesses both macroscopic morphological accuracy and rich microscopic detail, providing comprehensive three-dimensional morphological information for subsequent ablation risk assessment.

[0114] The above describes the fusion technique of microtexture and continuous depth mapping, as well as the structure boundary preservation method of anisotropic filtering. The fusion process of microtexture and depth mapping has complex technical implications: microtexture information comes from Laplacian edge enhancement processing and contains high-frequency information on surface detail changes; continuous depth mapping provides information on the macroscopic geometry and overall undulation of the surface; the purpose of fusion is to enhance the expressive ability of surface details while maintaining the accuracy of the overall morphology. The specific fusion algorithm adopts a multi-scale weight allocation strategy: in regions with drastic depth changes (such as ablation edges), the weight of microtexture information is higher to highlight boundary details; in regions with gentle depth changes, the weight of depth mapping information is higher to maintain the continuity of the morphology; the weight allocation is determined based on a combination of factors such as local gradient strength, curvature change, and texture complexity. Anisotropic filtering is a specialized edge-preserving smoothing technique. Its working principle is as follows: traditional isotropic filtering smooths uniformly in all directions, blurring important boundary information; while anisotropic filtering can identify the principal direction of the local structure, performing strong smoothing along the structure direction and maintaining sharpness perpendicular to the structure direction. Specifically, it determines the principal direction by calculating the local structure tensor: the eigenvectors of the structure tensor indicate the primary and secondary directions of the local gradient; the shape and orientation of the filter kernel are adaptively adjusted according to these principal directions, applying strong filtering in the boundary tangential direction and preserving the original information in the boundary normal direction. The final ablation depth map integrates all the above processing results: it contains accurate macroscopic geometric information, rich microscopic texture details, and clear structural boundaries; the map data format is a 3D point cloud combined with texture mapping, with each point containing spatial coordinates and texture attributes; the horizontal and vertical coordinates represent the two-dimensional position of the cable surface, and the depth coordinates represent the surface undulations caused by ablation; the texture attributes record information such as changes in material properties and the degree of ablation at that location. The technical problems solved by this high-precision ablation depth map include: providing a quantitative three-dimensional description of the ablation state and supporting the calculation of volume loss; establishing a visualization standard for ablation detection, which facilitates judgment and verification by professionals; and providing a detailed morphological benchmark for predicting the development trend of ablation.

[0115] For example, when thermal damage is detected in the outer sheath of a 110kV cable, the ablation classification engine first outputs the ablation classification results, dividing the cable surface into three parts: mild ablation, moderate ablation, and severe ablation, while also providing a confidence score for each region. Regions with confidence scores higher than 0.85 are selected as key detection intervals, and a 5mm buffer zone is extended outside the region boundaries to ensure complete capture of ablation variation characteristics. Subsequently, a composite sinusoidal phase-coded grating containing three sets of frequencies at 0.5mm... -1 1.0mm -1 and 2.0mm -1The sinusoidal fringes were used, with low-frequency fringes for coarse contour reconstruction of heavily ablated areas and high-frequency fringes for fine feature capture of lightly ablated areas. Reflection fringe images were acquired from four different angles, and the phase information of each angle was extracted using Fourier transform. The phase information from the four angles was then synthesized using a weighted average to reduce errors caused by shadows and reflections. The synthesized phase information was unwrapped using a Gauss-Helmholtz algorithm, with a regularization parameter λ set to 0.05 and 50 iterations, successfully eliminating phase jumps and constructing a continuous depth map. The depth map and ablated area markers were spatially registered using an iterative nearest-point algorithm, achieving sub-millimeter accuracy. The registered depth map was then processed using a Laplacian operator, employing a 5×5 Gauss-Laplacian kernel to highlight surface micro-texture, clearly showing the crack network in heavily ablated areas and the bubble distribution in moderately ablated areas. Finally, the Laplacian processing result was fused with the original depth map with a weight ratio of 7:3, and anisotropic filtering was performed for 15 iterations. The diffusion coefficient threshold was set to 20, generating a high-precision ablation depth map that retains the overall morphology and is rich in microscopic details, providing cable maintenance personnel with intuitive and clear ablation three-dimensional feature data.

[0116] In some embodiments, after determining the degree of erosion of the water-blocking buffer layer of the cable, the method further includes the following steps: obtaining evaluation index information of the water-blocking buffer layer of the cable, wherein the evaluation index information includes one or more of the following: degree of erosion, material strength, integrity of the water-blocking buffer layer, erosion rate, cable load, ambient temperature, and vibration stress; calculating a weighted average of the multiple evaluation index information to obtain an evaluation score; obtaining multiple standardized ranges, wherein the standardized ranges are predefined score intervals for the operational risk of the cable; determining the standardized range in which the evaluation score falls as the target standardized range, and determining the operational risk corresponding to the target standardized range as the current operational risk of the cable.

[0117] In this scheme, the multi-indicator comprehensive evaluation can fully reflect the health status of the cable's water-blocking buffer layer, avoiding the limitations of single-indicator evaluation. The calculation of the weighted average value takes into account the relative importance of different indicators in risk assessment, making the evaluation results more reasonable and reflecting the comprehensive level of cable operation risk.

[0118] Current risk assessment methods lack the ability to distinguish between thermal damage and mechanical damage, affecting the accuracy of the assessment. Specifically, the ablation depth map and the ablation grading engine results are integrated using a self-calibrating weighted fusion algorithm to calculate ablation risk quantification indicators. This includes: extracting depth ripple spectra, material embrittlement region maps, and microcrack distribution maps from the ablation depth map to construct a three-dimensional damage feature map; performing heterogeneous data registration between the three-dimensional damage feature map and the ablation grading engine results to establish a multi-level ablation characterization space; applying tensor decomposition technology to the multi-level ablation characterization space to separate thermal damage modes and mechanical damage modes, forming dual-modal damage components; processing the dual-modal damage components through a nonlinear self-calibrating network, dynamically adjusting parameters based on a historical case library to generate an optimized weight vector; adaptively fusing the three-dimensional damage feature map and the ablation grading engine results based on the optimized weight vector to construct an ablation damage integrity assessment system (i.e., assessment index information); calculating the severity of the current ablation state based on the ablation damage integrity assessment system, and outputting standardized ablation risk quantification indicators (i.e., the current operational risk of the cable).

[0119] Specifically, a three-dimensional damage feature map is constructed by extracting depth ripple spectra, material embrittlement region maps, and microcrack distribution maps from the ablation depth map. The depth ripple spectrum, a characteristic representation of the undulations of the ablated surface, is obtained through frequency domain analysis of the ablation depth map. Specifically, discrete wavelet transform is applied to the depth map to decompose the depth information into details and approximate components at different scales, and then the energy distribution at each scale is calculated to form a depth ripple spectrum characterizing the surface undulations. The material embrittlement region map identifies regions where the material's physical properties have changed by analyzing local curvature changes and texture features in the depth map. The identification process employs multi-threshold segmentation and morphological processing, combined with visual features of material embrittlement such as color changes and reduced surface smoothness, to accurately mark embrittlement regions of different degrees. The microcrack distribution map is a crack network extracted from the depth map using edge detection and linear structure enhancement algorithms. The algorithm first uses a directional filter bank to enhance the linear structure, then applies adaptive threshold segmentation and connectivity analysis to extract the complete crack network and record its spatial distribution, length, and orientation. These three feature images are combined to form a three-dimensional damage feature map, which comprehensively characterizes the physical damage state of the cable's water-blocking buffer layer.

[0120] Discrete wavelet transform is applied to the ablation depth map to decompose the depth information into details and approximate components at different scales. Then, the energy distribution at each scale is calculated to form a frequency domain description characterizing the surface undulations. The depth ripple spectrum reflects the undulation patterns of the ablated surface, including the spatial frequency, amplitude distribution, and directionality of the ripples. This spectrum can quantitatively describe the changes in surface roughness and micro-topographic features caused by ablation. The method for obtaining the material embrittlement region map is as follows: by analyzing local curvature changes and texture features in the depth map, regions where the material's physical properties have changed are identified. The system applies multi-threshold segmentation and morphological processing techniques, combined with visual characteristics of material embrittlement (such as color changes and reduced surface smoothness), to accurately mark embrittlement regions of different degrees. The embrittlement region map provides spatial distribution information on material strength degradation and is an important basis for assessing structural integrity. The generation of the microcrack distribution map employs a specialized linear structure detection algorithm: a directional filter bank is used to enhance linear structure features, highlighting the geometric characteristics of the cracks; then, adaptive threshold segmentation and connectivity analysis are applied to extract the complete crack network from the depth map; finally, the spatial distribution, length, orientation, density, and other geometric parameters of the cracks are recorded. The technical problems addressed by these three maps include: providing a multi-dimensional feature description of ablation damage, comprehensively characterizing the damage state from three perspectives: surface morphology, material properties, and structural integrity; establishing a hierarchical damage assessment system from macroscopic to microscopic levels; providing rich feature data for subsequent damage mode separation and risk quantification; and achieving a technological leap from qualitative analysis to quantitative assessment in ablation detection.

[0121] Heterogeneous data registration is performed between the 3D damage feature map and the ablation grading engine results to establish a multi-level ablation characterization space. Heterogeneous data registration refers to the process of spatially corresponding and semantically aligning data from different sources and with different representations. In cable ablation detection, it is necessary to establish an accurate correspondence between the 3D damage feature map of the physical morphology and the ablation level classification results output by the ablation grading engine. The registration process first establishes a coordinate transformation relationship, projecting the 3D damage feature map onto the same reference coordinate system as the grading engine results; then, fine alignment is performed through feature point matching, with matching points selected from significant feature locations of the ablation boundary; finally, non-rigid deformation adjustment is performed to compensate for local deformation caused by differences in acquisition angle and environmental conditions. After registration, a multi-level ablation characterization space is formed, including an ablation physical morphology layer, an ablation classification result layer, and an ablation development trend layer, each layer providing different dimensions of information on the ablation state.

[0122] The multi-level ablation characterization space is a comprehensive information framework integrating heterogeneous data from multiple sources. Its core function is to unify ablation information from different detection methods, data formats, and representation dimensions into a standardized spatial coordinate system, achieving organic fusion and collaborative analysis of information. The specific registration process includes several key steps: First, coordinate system unification, establishing a spatial correspondence between the three-dimensional damage feature map (containing three-dimensional spatial coordinates) and the ablation grading engine results (two-dimensional image coordinates). This process needs to address issues such as coordinate system differences, scale differences, and perspective differences between different data sources. All data is mapped to a unified world coordinate system through calibration matrices and geometric transformations. Second, time synchronization correction. Although it is a quasi-static detection, there are time intervals between different detection steps, requiring timestamp alignment and state consistency verification to ensure data timeliness. Third, feature point matching and fine alignment. The system selects significant feature points of the ablation boundary as registration benchmarks and uses an iterative nearest-point algorithm for sub-pixel-level precise alignment. Finally, non-rigid deformation correction. Considering the minute deformations that may occur in the cable at different detection stages, deformation compensation is performed using methods such as thin-plate spline interpolation. The multi-layered representation space formed after registration has the following structural features: the first layer is the physical morphology layer, containing geometric and material information such as depth ripple spectrum, embrittlement region map, and crack distribution map; the second layer is the classification result layer, containing the degree assessment and confidence information output by the ablation grading engine; and the third layer is the development trend layer, which is the ablation evolution prediction inferred based on historical data and physical models. The registration mentioned does indeed integrate multiple datasets into a unified dataset, but this is not a simple data stacking. Rather, it establishes spatial and semantic relationships between them while maintaining the independence of each data source, forming a multi-dimensional, multi-layered, and interactively queried comprehensive information space.

[0123] Tensor decomposition (TFD) was applied to the multi-level ablation characterization space to separate thermal and mechanical damage modes, forming a bimodal damage component. TFD is a multidimensional data analysis method that decomposes high-dimensional data into a combination of a low-dimensional core tensor and factor matrices. In cable ablation detection, the multi-level ablation characterization space was organized as a third-order tensor, with the three dimensions representing spatial location, feature type, and data source, respectively. Tucker decomposition was applied to process this tensor, decomposing it into a core tensor and the product of the three factor matrices. By analyzing the characteristic patterns of the factor matrices, the main modes related to thermal and mechanical damage were identified. Thermal damage modes mainly manifest as uniform ablation of the material surface, color changes, and material softening, while mechanical damage modes mainly manifest as non-thermal damage forms such as cracks, scratches, and indentations. The separated thermal and mechanical damage modes form a bimodal damage component, providing a more refined distinction between damage types for subsequent risk assessment.

[0124] The initial separation was primarily based on differences in spectral characteristics, distinguishing different types of damage responses in the frequency domain. This separation was relatively coarse, mainly to optimize subsequent image processing. The aforementioned separation, however, is a refined separation based on richer physical features after obtaining complete three-dimensional morphology and material property information, aiming to provide accurate damage pattern classification for risk quantification. The specific separation method employs tensor decomposition technology: the multi-level ablation characterization space is organized into a high-order tensor, where each dimension represents spatial location, feature type, data source, etc.; then, Tucker decomposition is applied to decompose this high-dimensional tensor into a core tensor and the product of multiple factor matrices; by analyzing the characteristic patterns of the factor matrices, the system can identify the main patterns related to thermal and mechanical damage. The judgment conditions and parameter settings are based on the physical mechanism of ablation: thermal damage modes mainly manifest as changes in the uniformity of materials, such as softening, carbonization, and melting. In the characteristic space, these are characterized by the dominance of low-frequency components, relatively continuous spatial distribution, and a high correlation with temperature history. Mechanical damage modes mainly manifest as structural damage caused by external forces, such as cracks, scratches, and indentations. These are characterized by prominent high-frequency components, linear or point-like spatial distribution, and correlation with stress concentration areas. Specific discrimination parameters include: frequency domain energy ratio thresholds (thermal damage Rthermal > 0.6, mechanical damage Rmechanical > 0.4), spatial connectivity indices (high connectivity for thermal damage, low connectivity for mechanical damage), and morphological characteristics (smooth boundaries for thermal damage, sharp boundaries for mechanical damage). The technical value of this separation lies in: providing more accurate damage type identification, allowing for different risk assessment models for different types of damage; establishing a quantitative relationship between damage causes and damage manifestations; and providing a scientific basis for the formulation of maintenance strategies, as different damage types require different treatment methods.

[0125] The dual-modal damage components are processed through a nonlinear self-calibrating network, and parameters are dynamically adjusted based on a historical case database to generate an optimized weight vector. A nonlinear self-calibrating network is a neural network structure capable of automatically adjusting its internal parameters based on historical data. In the cable ablation detection system, this network receives the dual-modal damage components as input and dynamically adjusts its parameters by comparing them with samples in the historical case database. The historical case database contains a large amount of cable sample data with known ablation degrees and subsequent development results; each sample includes information such as damage characteristics, actual risk level, and evolution history. The network learning process employs an adaptive learning rate strategy, dynamically adjusting the learning step size based on the similarity between the current sample and historical cases, assigning more learning weights to less common damage patterns. After processing, the network outputs an optimized weight vector, representing the contribution of different damage features to the overall risk, providing guidance for subsequent adaptive fusion.

[0126] The parameter weights are the core output of the nonlinear self-calibrating network, and each weight value has a clear physical meaning and mathematical definition. These weight parameters correspond to the contribution of each feature dimension in the dual-modal damage component to the overall risk assessment. Specifically, the dimensions of the weight vector correspond to the dimensions of the feature space, and typically include: thermal damage feature weighting, including weights for material softening, surface carbonization, and thermal deformation, with each weight value between 0 and 1, representing the degree of influence of the feature on thermal damage risk; mechanical damage feature weighting, including weights for crack density, crack depth, surface scratches, and indentation deformation, similarly representing the contribution of each feature to mechanical damage risk; and composite damage interaction weights, reflecting the synergistic effect between thermal and mechanical damage, because in reality, the two types of damage often influence and exacerbate each other. The numerical calculation of weights is based on statistical analysis of a historical case database: the system collects a large number of cable erosion samples with known risk levels and analyzes the correlation between different features and the final risk outcome through machine learning algorithms; the dynamic adjustment mechanism of weights considers the similarity between the current sample and historical cases, with historical cases with high similarity having a greater impact on weight calculation; the self-correction process continuously optimizes weight allocation through feedback learning, and when the predicted results deviate from the actual results, the system automatically adjusts the relevant weights to improve prediction accuracy. The technical significance of these weights lies in: quantifying the actual impact of different damage characteristics on overall risk, avoiding the bias of subjective judgment; providing an interpretable risk assessment process, allowing experts to understand the specific contribution of each factor; establishing a precise mapping relationship from features to risk, realizing the standardization and automation of risk assessment; and supporting personalized weight adjustments, which can be optimized according to different application scenarios and risk preferences.

[0127] An adaptive fusion of the 3D damage feature map and the ablation grading engine results, based on an optimized weight vector, is used to construct an ablation damage integrity assessment system. Adaptive fusion refers to dynamically adjusting the weights of different features during the fusion process according to their importance and reliability, to obtain a more accurate comprehensive assessment result. In cable ablation detection, the fusion process first assigns initial weights to each feature layer in the 3D damage feature map and the classification results of the ablation grading engine; then, these initial weights are adjusted according to the optimized weight vector, increasing the weight of key features and decreasing the weight of noisy or unstable features; finally, the fusion result is generated through weighted summation or weighted voting. The fusion result constitutes a multi-dimensional ablation damage integrity assessment system, including four dimensions: ablation depth assessment, material strength assessment, structural integrity assessment, and development trend assessment, comprehensively characterizing the ablation damage state of the cable's water-blocking buffer layer from different perspectives.

[0128] The technical logic of weighted fusion is based on the theory of multi-source information fusion. Its core idea is that information provided by different data sources is complementary and redundant. Through appropriate fusion, more accurate and reliable results can be obtained than those from a single data source. The specific fusion process includes: performing information consistency checks to verify the consistency between the 3D damage feature map and the ablation grading engine results in terms of spatial location and damage degree judgment. If significant inconsistencies are found, the system will initiate a data quality check and recalibration process. Then, confidence assessment is performed, calculating the reliability indices of the two data sources respectively. The confidence of the 3D morphology data is based on factors such as measurement accuracy, surface conditions, and illumination quality, while the confidence of the grading engine results is based on model training quality, feature matching degree, and historical validation results. Next, adaptive weight allocation is performed. The system dynamically adjusts the weights of the two data sources according to the current detection conditions. When surface conditions are good and 3D reconstruction accuracy is high, higher weights are given to the morphology data. When image features are obvious and classification confidence is high, higher weights are given to the grading results. While the construction of the assessment system does exhibit some characteristics of being manually defined, it is based on extensive engineering practice and scientific research, possessing a solid theoretical foundation: the system's framework references equipment integrity assessment methods in international standards, integrating knowledge from multiple disciplines such as materials science, structural engineering, and reliability engineering; the selection of assessment indicators is based on the physical mechanisms of cable ablation and engineering experience, with each indicator having clear physical meaning and engineering value; the determination of weighting coefficients is derived through statistical analysis of numerous historical cases, and has undergone expert review and on-site verification; the system's effectiveness has been verified through comparison with traditional testing methods, demonstrating significant advantages in accuracy, consistency, and repeatability. The rationale for this manually defined approach lies in: transforming complex physical phenomena into quantifiable mathematical models, turning subjective engineering judgments into objective numerical calculations; providing a standardized assessment process, eliminating potential subjective differences arising from different testing personnel or testing times; and establishing traceable and verifiable assessment records, facilitating quality control and accountability.

[0129] The severity of the current ablation state is calculated based on the ablation damage integrity assessment system, outputting a standardized ablation risk quantification index. The severity calculation employs a multi-dimensional risk assessment model, mapping various indicators in the ablation damage integrity assessment system to a unified risk space. The calculation process first determines the risk contribution function for each dimension, describing the correspondence between the indicator value and the risk level; then, the risk contributions of each dimension are weighted and combined according to their importance in the cable system; finally, risk normalization is performed, mapping the risk values ​​to a standard range of 0-100, forming a standardized ablation risk quantification index that is easy to understand and compare. This index considers both the current state of ablation and its potential development trend, providing an intuitive and clear basis for cable maintenance decisions.

[0130] The calculation of individual risk indicators is based on various dimensions (depth assessment, strength assessment, integrity assessment, and trend assessment) in the ablation damage integrity assessment system, calculating the risk contribution value for each dimension separately. The depth assessment risk (Rdepth) is calculated based on the ratio of ablation depth to the cable design safety margin; the risk increases sharply when the ablation depth exceeds the safety threshold. The strength assessment risk (Rstrength) is calculated based on the degree of material embrittlement and remaining load-bearing capacity, considering the impact of material strength degradation on cable performance. The integrity assessment risk (Rintegrity) is calculated based on the development degree and connectivity of the crack network to assess the degree of structural integrity damage. The trend assessment risk (Rtrend) is calculated based on the ablation development rate and predicted evolution trajectory to assess the future risk development trend. Then, a comprehensive calculation of multi-dimensional risks is performed using a weighted geometric average method: Rtotal = (Rdepth) / (Rtenderness ... w1 ×Rstrength w2 ×Rintegrity w3 ×Rtrend w4 ) (1 / (w1+w2+w3+w4)) The coefficients w1, w2, w3, and w4 are weighting coefficients for each dimension, determined based on factors such as cable type, operating environment, and importance. The final determination of the hazard level also considers the influence of environmental factors, including the amplification or mitigation effects of external conditions such as cable load level, ambient temperature and humidity, and vibration stress on the ablation risk. The system has established an environmental impact factor database, allowing for the lookup of corresponding correction coefficients based on current operating conditions. The final standardized risk index uses linear mapping to convert the calculation results to a standard range of 0-100 for easy understanding and comparison. The output standardized ablation risk quantification index has clear engineering significance: 0-30 represents a low-risk zone, indicating slight ablation and continued normal operation; 30-60 represents a medium-risk zone, requiring enhanced monitoring and a preventative maintenance plan; 60-80 represents a high-risk zone, recommending prompt repair or replacement; and 80-100 represents an extremely high-risk zone, posing a serious safety hazard and requiring immediate shutdown. The technical value of this calculation process lies in the fact that it provides a scientific, objective, and repeatable method for risk quantification; it establishes an effective conversion from complex physical phenomena to simple numerical indicators; it provides clear numerical basis for engineering decisions; and it realizes the standardization and automation of risk assessment.

[0131] For example, the detection system first extracts feature information from the ablation depth map obtained in the previous stage. By extracting the depth ripple spectrum through wavelet decomposition, it was found that the frequency at 2.5 cm⁻¹... -1 up to 5cm -1An abnormal concentration of energy within the range indicates regular ripple changes on the surface, which is usually related to thermal stress cycling. A multi-threshold segmentation algorithm was applied to identify embrittlement regions in the material, revealing a severely embrittled zone approximately 15 mm in diameter in the center of the water-blocking buffer layer, surrounded by a transitional embrittlement zone. Directional filtering and linear structure enhancement algorithms were used to extract the microcrack network, detecting crack clusters mainly distributed radially, with the longest crack reaching 7.8 mm, and multiple cracks exhibiting an intersecting network structure. These three types of feature information combined to form a complete three-dimensional damage feature map. This map was then registered with the classification results output by the ablation grading engine, using 12 feature point matching and non-rigid deformation correction to achieve precise spatial correspondence. Tucker tensor decomposition was applied to the registered multi-level characterization space, extracting three main components: two related to thermal damage modes, manifested as uniform material softening and surface carbonization; and one related to mechanical damage modes, manifested as cracks and deformation caused by external forces. These components are input into a nonlinear self-calibrating network trained on 2500 historical cases. The network dynamically adjusts the processing parameters based on the current sample characteristics and outputs a 15-dimensional optimized weight vector. The weight of thermal damage features is generally higher than that of mechanical damage features, indicating that the current sample is mainly affected by thermal effects. Based on this weight vector, the system performs weighted fusion of various features to construct a complete evaluation system including depth assessment, strength assessment, integrity assessment, and trend assessment. The final calculated standardized risk index is 78.5, belonging to the "high-risk" level. The main risk contributions come from the continuous high-temperature effects in the material embrittlement region and the potential propagation of the crack network. The system also provides a maintenance recommendation of "replacement within 72 hours," providing cable maintenance personnel with a clear decision-making basis.

[0132] Current predictive and early warning mechanisms are inadequate, making it difficult to accurately predict and promptly warn of ablation trends. Specifically, based on the aforementioned ablation risk quantification indicators, a time-series predictive extrapolation mechanism is used to generate a distributed visualization report and early warning threshold recommendations. This includes: performing time-series decomposition on the aforementioned ablation risk quantification indicators to separate long-term trends, periodic changes, and random disturbances, constructing an ablation evolution pattern library; inputting the ablation evolution pattern library into a recursive waveform recognizer to extract typical degradation path features and establish an ablation propagation prediction baseline; applying multi-scale entropy analysis to the aforementioned ablation propagation prediction baseline to determine the critical transition points of the stable zone, accelerated degradation zone, and danger zone, forming an ablation early warning threshold spectrum; converting the aforementioned ablation early warning threshold spectrum into a color-coded interactive layer, overlaying it onto the cable structure's 3D model to generate a three-dimensional visualization interface; adding a time axis control component to the aforementioned three-dimensional visualization interface to achieve ablation state retrospective and forward-looking extrapolation functions, constructing an early warning decision support system; and integrating the data analysis and prediction results generated by the aforementioned early warning decision support system into a standardized report format, outputting a distributed visualization report and early warning threshold recommendations.

[0133] Specifically, time series decomposition was performed on the quantitative indicators of ablation risk to separate long-term trends, periodic changes, and random disturbances, thus constructing an ablation evolution pattern library. Time series decomposition is an analytical method that breaks down time series data into different components to identify inherent patterns within the data. In cable ablation monitoring, the continuously collected quantitative indicators of ablation risk were subjected to seasonal trend decomposition. The STL decomposition method was used to split the time series into three main components: the long-term trend reflects the overall direction of ablation severity over time, typically showing a slow increase or a stable state; periodic changes reflect the regular fluctuations in ablation indicators caused by factors such as cable load cycles and changes in environmental temperature and humidity; and the random disturbance component includes irregular changes caused by power grid fluctuations, sudden overloads, etc. After feature extraction, these three components were stored in the ablation evolution pattern library. Each pattern contains attribute information such as time span, rate of change, and sensitive factors, providing basic data support for subsequent predictive analysis.

[0134] The three identified trends have different physical meanings and technical values: The long-term trend reflects the overall development direction of the ablation state, representing the natural aging and cumulative damage of cable materials during long-term operation. It typically exhibits a slow upward trend or relative stability, with a time span potentially reaching months or even years. This trend provides an important basis for cable lifecycle management and long-term maintenance planning. Periodic changes reflect fluctuations in ablation indicators caused by regular factors, mainly including daily load cycles (load differences between weekdays and rest days), seasonal cycles (the impact of temperature and humidity changes on material performance), and annual maintenance cycles. Identifying these periodic changes helps distinguish between normal operational fluctuations and abnormal ablation development, avoiding misjudging normal fluctuations as escalating risks. The random disturbance component includes changes in ablation indicators caused by unpredictable factors such as sudden events, environmental anomalies, and operational errors, such as lightning strikes, overloads, and mechanical shocks. Although these disturbances are random, statistical analysis can assess their impact on overall risk. The specific decomposition technique employed is the STL decomposition method (Seasonal and Trend decomposition using Loess), a robust time series decomposition method capable of handling nonlinear trends and time-varying periodicity. The technical challenges addressed by this decomposition process include: providing a cleaner and more targeted data foundation for subsequent predictive modeling; helping to identify the inherent patterns of ablation development and distinguish between predictable changes and random disturbances; establishing a risk evolution pattern library to provide a reference for risk prediction of similar cables; and improving the accuracy of anomaly detection, enabling timely identification of risk changes deviating from normal patterns.

[0135] The ablation evolution pattern library is input into a recursive waveform recognizer to extract typical degradation path features and establish an ablation propagation prediction baseline. The recursive waveform recognizer is a time series analysis tool based on recurrent neural networks, capable of identifying specific patterns in time series data and predicting their future trends. In cable ablation prediction, the recursive waveform recognizer employs a long short-term memory network structure, using a gating mechanism to handle long-term dependencies, and learns to identify typical evolution paths for different ablation types from the ablation evolution pattern library. The recognition process first converts the time series data of historical ablation cases into feature sequences, including the rate of change of indicators, acceleration, and coefficient of variation; then, the network is trained to identify feature sequence templates for different ablation modes from these feature sequences; finally, the currently monitored ablation indicators are matched with the templates to find the most similar degradation path pattern. Based on the matching results and the current state, the system estimates the predicted values ​​of ablation indicators at future time points, forming an ablation propagation prediction baseline that describes the possible development trajectory of the cable ablation state.

[0136] Historical data from the ablation evolution pattern library is converted into feature sequences, each containing time-series features such as the rate of change of indicators, acceleration, and coefficient of variation. Then, a Long Short-Term Memory (LSTM) network structure is employed, using a gating mechanism to handle long-term dependencies and learn typical evolution paths for different ablation patterns. The recognizer can extract representative degradation patterns from complex time-series data, such as "slow degradation," "accelerated degradation," and "sudden degradation." The establishment of the ablation propagation prediction baseline is more complex: the system matches the current cable's ablation index time series with the typical degradation paths learned by the recognizer to find the most similar historical patterns. Then, based on the evolution trajectory of similar patterns and combined with the current cable's specific operating conditions (load level, environmental conditions, material properties, etc.), a personalized prediction baseline is established. This baseline is not a simple numerical extrapolation but a comprehensive prediction model that considers physical mechanisms and statistical laws. The technical issues addressed by establishing a baseline include: providing a scientific basis for risk prediction, avoiding the arbitrariness of subjective judgment; establishing personalized prediction models that consider the specific characteristics of different cables; achieving an effective conversion from historical experience to future predictions; and providing a quantitative basis for selecting the timing of maintenance decisions. The technical value of a prediction baseline lies in: its ability to predict the ablation trend over the next 3-6 months, providing a time window for maintenance planning; identifying key risk inflection points and providing early warnings of potential rapid deterioration; and supporting the evaluation of the effectiveness of different maintenance strategies, helping to select the optimal maintenance solution.

[0137] Multi-scale entropy analysis is applied to the ablation propagation prediction baseline to determine the critical transition points of the stable zone, accelerated deterioration zone, and danger zone, forming an ablation early warning threshold spectrum. Multi-scale entropy analysis is a method for assessing the complexity of time series data, reflecting the uncertainty and complexity of data at different scales. In cable ablation prediction, multi-scale entropy analysis first reconstructs the prediction baseline at multiple scales, calculating sample entropy values ​​at different time granularities; then, it analyzes the entropy value change trend, identifying abrupt change points and inflection points; finally, combining the entropy value change characteristics and the physical evolution law of ablation, it determines the key state transition points in the ablation development process. Generally, the stable zone corresponds to an area with low and gradual entropy values, indicating a relatively stable ablation state; the accelerated deterioration zone corresponds to an area where the entropy value begins to rise significantly, indicating that the ablation state begins to deteriorate rapidly; the danger zone corresponds to an area where the entropy value reaches its peak and then begins to fluctuate or decline, indicating that the ablation has reached a critical state and there is a serious safety hazard. The index values ​​of these boundary points constitute the ablation early warning threshold spectrum, setting clear early warning trigger conditions for different degrees of ablation.

[0138] Multi-scale entropy analysis reconstructs the ablation propagation prediction baseline across multiple time scales, calculating sample entropy values ​​from fine-grained (hourly) to coarse-grained (monthly) scales. Sample entropy reflects the complexity and uncertainty of the time series, and the changing patterns of entropy values ​​can reveal the transition characteristics of the system state. The stable zone is identified as an area where entropy values ​​remain relatively low and change slowly across multiple time scales, corresponding to a risk index below approximately 40. This indicates a relatively stable ablation state with slow development, capable of maintaining normal operation. Within this zone, ablation mainly manifests as natural aging of the material, without a significant trend of accelerated deterioration. The accelerated deterioration zone is characterized by a significant increase in entropy values, particularly at medium time scales (days to weeks), corresponding to a risk index between approximately 40 and 70. This indicates that ablation has entered an accelerated development stage, possibly due to increased load, environmental degradation, or accumulated damage reaching a critical point. Close monitoring and preventative maintenance plans are required in this zone. The criteria for determining the danger zone are: after the entropy value reaches its peak, it begins to fluctuate or decrease, and the corresponding risk index exceeds 70, indicating that the system has entered a highly unstable state and there is a risk of sudden failure. A decrease in entropy value does not necessarily mean a reduction in risk, but rather that the system's change pattern has become more deterministic, usually foreshadowing severe structural damage. The determination of critical transition points employs a change point detection algorithm, using statistical testing methods to identify significant changes in the entropy value sequence. These transition points have significant early warning value and are key time nodes for making maintenance decisions. The resulting ablation early warning threshold spectrum includes: a green warning threshold (upper limit of the stable zone, approximately 40), a yellow warning threshold (upper limit of the accelerated degradation zone, approximately 70), and a red warning threshold (lower limit of the danger zone, approximately above 70). This threshold system provides clear numerical boundaries for different risk levels, achieving the quantification and standardization of risk management.

[0139] The ablation warning threshold spectrum is converted into a color-coded interactive layer and overlaid on the 3D model of the cable structure to generate a three-dimensional visualization interface. Color coding is a method of converting numerical information into an intuitive visual representation, reflecting the magnitude or category of data values ​​through color changes. In cable ablation visualization, the threshold spectrum is first divided into multiple intervals, establishing a mapping relationship between index values ​​and colors. A gradient color spectrum from green to yellow to red is typically used, with green representing safe areas, yellow representing warning areas, and red representing danger areas. Then, based on the spatial distribution of ablation risk indicators on the cable structure, the corresponding color information is mapped onto the surface of the 3D cable model, forming a risk distribution heatmap covering the entire cable. Simultaneously, the system adds interactive markers to key risk points, allowing users to click to view detailed information or initiate in-depth analysis. The 3D cable structure model is built based on high-precision CAD data, including complete structural information such as the outer sheath, water-blocking buffer layer, insulation layer, and conductor. Semi-transparent rendering technology enables simultaneous visualization of multiple layers, allowing for a direct display of the internal ablation state. The resulting three-dimensional visualization interface provides both a macroscopic panoramic view of the ablation risk distribution and supports microscopic detail viewing of key areas.

[0140] The ablation warning threshold spectrum is color-coded according to risk level, using internationally recognized risk color standards: green represents safe areas (risk index 0-40), yellow represents warning areas (40-70), orange represents dangerous areas (70-85), and red represents extremely dangerous areas (85-100). This color information is then used as texture maps and mapped onto the 3D geometric model of the cable structure. The 3D model of the cable structure is built based on high-precision CAD data and includes a complete hierarchical structure such as the outer sheath, water-blocking buffer layer, insulation layer, and conductor. The model uses semi-transparent rendering technology, allowing visualization of the internal structure and enabling users to view the ablation status at different levels. The technical features of the 3D display include: support for 360-degree rotation and multi-angle observation, allowing users to check the cable status from any perspective; multi-level zoom functionality, providing clear display from overall overview to local details; real-time interactive operation, allowing users to click on any location to view detailed risk data; and integrated timeline control, which can display the historical evolution of risk distribution. The technical problems solved by this 3D display include: converting abstract numerical information into intuitive visual expressions, greatly improving the comprehensibility of information; providing a spatialized risk distribution display, making it easier to identify risk concentration areas and development trends; realizing the comprehensive display of multi-dimensional information, integrating multiple information such as location, degree, and time in one interface; and enhancing the user experience, enabling professional test results to be understood and applied by a wider range of users.

[0141] A timeline control component was added to the 3D visualization interface to enable backtracking and forward-looking simulations of ablation status, thus constructing an early warning decision support system. Timeline control is an interactive data exploration method that allows users to move freely along the time dimension and observe the changes in data over time. In the cable ablation monitoring system, the timeline control component includes elements such as sliders, play buttons, and time point selectors, allowing users to flexibly switch between historical records and future predictions. The backtracking function allows viewing ablation status data at any historical point in time, including measured indicators, 3D morphology, and risk assessment results, helping to analyze the development process and causes of ablation. The forward-looking simulation function, based on the aforementioned prediction baseline, displays the predicted ablation status at different future points in time and shows the risk change trends under different prediction scenarios. The system also provides a hypothesis analysis tool, allowing users to adjust operating parameters or maintenance strategies and observe the impact of different decisions on ablation risk. These functions together constitute the early warning decision support system, providing comprehensive decision assistance to cable maintenance personnel.

[0142] The "Timeline Control Component" is an interactive data exploration tool whose core function is to allow users to freely browse and analyze the evolution of ablation states over time. Specific control components include: a time slider, allowing users to drag the slider between historical records and future predictions to view the cable status at a specific point in time; a playback control button, supporting automatic playback of the ablation evolution animation with adjustable speed; a time jump input box, allowing users to directly input specific dates and times for quick location; and a time range selector, allowing users to select and view data changes within a specific time period. The backtracking function is implemented as follows: the system maintains complete historical data records, including raw data, processing results, risk assessments, and other information for each detection; after the user selects a historical time point, the system automatically loads the corresponding data and updates the 3D display, allowing users to see the ablation distribution, risk level, and detection results at that moment; the backtracking function also supports comparative analysis of multiple time points, allowing users to simultaneously view the status of different periods and observe the development and changes in ablation. The forward-looking simulation function is based on the previously established predictive model: the system calculates possible states at different future points in time based on the current state and historical trends; users can view the predicted ablation trajectory to understand the possible evolution direction of the risk; the simulation results contain uncertainty information, displaying the reliability range of the prediction through confidence intervals and other methods; users can also simulate different maintenance strategies and observe their impact on future risk development. The early warning decision support system built upon these control functions has the following value: it provides rich time-dimensional information for maintenance decisions, helping to determine the optimal maintenance timing; it supports root cause analysis, allowing for tracing the origin and development process of risk problems by reviewing historical data; and it provides quantitative evidence for preventative maintenance, enabling the development of forward-looking maintenance plans based on the prediction results.

[0143] The data analysis and prediction results generated by the early warning decision support system are integrated into a standardized report format, outputting distributed visualization reports and early warning threshold recommendations. Standardized reports transform complex analytical results into structured, easily understood, and disseminated information. In cable erosion monitoring, the report generation process first extracts key result data from the early warning decision support system, including current erosion status assessment, development trend prediction, and risk level determination. This data is then organized into a predefined report template, comprising four main parts: summary, detailed analysis, early warning recommendations, and maintenance plan. Finally, the report's content depth and presentation format are automatically adjusted based on user roles and terminal device types, ensuring that users at different levels can access information tailored to their needs. Distributed visualization means the report can be displayed simultaneously on multiple platforms and terminals, including control center screens, engineer workstations, and field mobile devices, with all terminals maintaining real-time data synchronization. The early warning threshold recommendation section, based on historical data analysis and current monitoring results, provides personalized threshold setting suggestions for monitoring cable water-blocking buffer layer erosion risks under different operating conditions, ensuring the early warning system can identify potential risks promptly and accurately.

[0144] The standardized report generation and early warning threshold recommendations described above are the final output of the entire detection system, and their implementation involves complex data integration and intelligent analysis technologies. The specific report generation process is as follows: key analytical results are extracted from the early warning decision support system, including core information such as current ablation status assessment, development trend prediction, risk level determination, and identification of key risk points; then, this technical data is organized according to a predefined report template, which includes four main parts: an executive summary, detailed analysis, early warning recommendations, and a maintenance plan. The executive summary uses concise language to summarize key findings and main recommendations, targeting management users; the detailed analysis includes complete technical data, charts, calculation processes, etc., targeting technical personnel; the early warning recommendations provide specific operational suggestions based on the risk analysis results; and the maintenance plan combines the prediction results to develop a phased maintenance plan. Personalized report generation considers differences in user roles and terminal devices: for management users, the report highlights key information such as risk level, impact degree, and recommended measures; for technical personnel, the report provides detailed technical parameters, analysis processes, data charts, etc.; for field operators, the report focuses on specific inspection locations, operational requirements, and safety precautions. Distributed visualization is achieved through multi-platform synchronization technology: reports can be simultaneously displayed on multiple terminals such as the control center's large screen, engineer workstations, and field mobile devices; all terminals maintain real-time data synchronization to ensure information consistency; and the display format is automatically adjusted according to the screen size and interaction method of different terminals. The generation of early warning threshold suggestions is based on historical data analysis and current monitoring results: the system analyzes the cable erosion development patterns under different operating conditions and provides personalized threshold setting suggestions for specific cable types, operating environments, and load conditions; the suggestions include detailed information such as the specific values ​​of multi-level early warning thresholds, triggering conditions, and response measures; the threshold suggestions also consider the balance between false alarm rate and false negative rate, determining the optimal threshold setting through ROC curve analysis; the system also provides a dynamic threshold adjustment mechanism, continuously optimizing threshold settings based on actual operating experience and feedback from detection results. The technical value of this report and suggestion generation mechanism lies in: achieving an effective conversion from complex technical data to concise decision-making information; providing a standardized and regulated detection report format for easy archiving and comparative analysis; and establishing a closed-loop quality management system that continuously improves the accuracy of detection and early warning through continuous feedback and optimization.

[0145] Taking a 220kV cross-linked polyethylene cable in a power transmission network as an example, after the monitoring system continuously collected three months of ablation risk quantification data, it first performed time series decomposition on these data. The decomposition results showed that the long-term trend was a slow upward trend, with the risk index baseline increasing from the initial 27.3 to 32.8 within three months; the periodic changes showed obvious daily and weekly load cycles, with the risk index during peak weekdays being 3.5-5.2 points higher than on rest days; random disturbance component analysis revealed two significant abnormal fluctuations, corresponding to the lightning strike on July 15 and the temporary overload operation on August 23, respectively. These decomposed components were stored in an ablation evolution pattern library, forming the characteristic evolution pattern of this type of cable. This pattern library was input into a recursive waveform recognizer trained on 5000 historical cases. The system identified that the current cable ablation pattern was closest to "slow degradation caused by thermal cycling," predicting that the risk index would continue to rise slowly over the next six months, with an expected growth rate of 1.2-1.8 points per month. Multi-scale entropy analysis was applied to the predicted baseline, calculating sample entropy at 10 different time scales. The results showed a significant increase in entropy at the fourth scale, corresponding to a risk index of approximately 42.5, which was identified as the transition point from the stable zone to the accelerated degradation zone. At the seventh scale, the entropy reached its peak and then began to decline, corresponding to a risk index of approximately 67.8, which was identified as the transition point from the accelerated degradation zone to the danger zone. Based on these analysis results, the system constructed a threshold spectrum with three levels of warnings: a risk index of 40-60 indicates a yellow warning (observation is advised), 60-75 indicates an orange warning (close monitoring), and above 75 indicates a red warning (emergency action is required). The system converted this warning spectrum into a gradient color from green to red, superimposed it onto the cable's 3D model, clearly showing that the intermediate joint area was approaching a yellow warning state. Users could advance the timeline by three months using the time axis control component, observing that the cable intermediate joint area turned yellow, while six months prior, some areas had turned orange, indicating that without intervention, the cable would enter a high-risk state within six months. The standardized report generated by the system incorporates these analytical findings and recommends preventative maintenance within four months, including measures such as reducing operating temperature and increasing the frequency of infrared inspections. The report is simultaneously pushed to the power grid control center and field maintenance teams via a distributed system, providing data support for operational and maintenance decisions.

[0146] The technical solution provided in this application employs a dual-spectrum imaging device to simultaneously acquire visible and infrared light data of the cable's water-blocking buffer layer. After nonlinear contrast optimization processing, an enhanced dual-spectrum image is obtained, significantly improving the recognizability of ablation features and solving the problem of traditional single-band imaging's difficulty in simultaneously capturing surface texture and thermal distribution features. The enhanced dual-spectrum image is then used to locate the ablation region through an adaptive region growing segmentation method combined with edge-preserving filtering. The texture gradient feature matrix is ​​extracted to obtain the ablation feature spectrum, achieving accurate segmentation and feature extraction of ablation regions under complex backgrounds. This overcomes the limitations of traditional fixed-threshold segmentation in uneven lighting and complex surfaces. Limitations under surface conditions; based on the ablation feature spectrum, an improved residual attention network is used to establish an ablation degree assessment model, which is trained through transfer learning to form an ablation grading engine. This network structure fully considers the hierarchy and correlation of cable ablation features, highlights key features through the attention mechanism, solves the problem of training deep networks through residual connections, and effectively solves the problem of insufficient samples through transfer learning, enabling the model to accurately distinguish different degrees of ablation. Based on the output of the ablation grading engine, the micro-morphology of the ablated surface is constructed using phase structured light coding technology to generate a high-precision ablation depth map, realizing the three-dimensional reconstruction of the micro-morphology of cable surface ablation. The system provides quantitative morphological parameters for damage assessment. By integrating ablation depth maps with ablation classification engine results using a self-calibrating weighted fusion algorithm, a quantitative ablation risk index is calculated. This achieves the organic fusion of physical morphological features and intelligent classification results. The self-calibrating mechanism dynamically adjusts the fusion weights, improving the accuracy and robustness of risk assessment. Based on the quantitative ablation risk index, a time-series prediction and extrapolation mechanism is used to generate distributed visualization reports and early warning threshold suggestions, constructing a complete closed loop from detection to prediction. This not only achieves accurate assessment of the current ablation state but also predicts its future development trend and provides targeted early warnings and treatment suggestions. This significantly improves the foresight and decision-making efficiency of cable maintenance. The artificial intelligence algorithms in this solution make a significant contribution to the overall solution, especially the application of the improved residual attention network in ablation classification. Through the synergistic effect of residual connections and attention mechanisms, it effectively solves problems such as complex cable ablation characteristics and imbalanced samples, achieving high-precision classification. The recursive waveform recognizer and multi-scale entropy analysis algorithm in the time-series prediction and inference mechanism have been specifically optimized for the time-series characteristics of cable ablation, making the prediction results more consistent with the actual ablation evolution law. These algorithm features are deeply integrated with the application scenarios of cable ablation detection, improving detection efficiency and accuracy.

[0147] In summary, the above methods, through multi-source data fusion and deep learning technology, construct a complete technical path from ablation feature extraction, three-dimensional morphology reconstruction, intelligent hierarchical assessment to risk quantification and early warning, thereby improving the accuracy, comprehensiveness and foresight of cable ablation detection.

[0148] This application also provides a device for detecting the degree of erosion of a cable water-blocking buffer layer. It should be noted that this device can be used to execute the method for detecting the degree of erosion of a cable water-blocking buffer layer provided in this application. This device is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0149] The following describes the cable water-blocking buffer layer ablation detection device provided in the embodiments of this application.

[0150] Figure 4 This is a structural block diagram of a cable water-blocking buffer layer ablation detection device according to an embodiment of this application. Figure 4 As shown, the device includes:

[0151] The first acquisition unit 10 is used to acquire visible light and infrared images of the water-blocking buffer layer of the cable;

[0152] The first fusion unit 20 is used to perform weighted fusion of the above-mentioned visible light image and the above-mentioned infrared image to obtain the target image;

[0153] The decomposition unit 30 is used to perform multi-scale decomposition on the target image to obtain multiple spatial frequency images corresponding to the target image, wherein the multi-scale decomposition is to decompose the image into different spatial frequencies.

[0154] The first determining unit 40 is used to determine the ablation of the water-blocking buffer layer of the cable when the spatial frequency corresponding to at least one of the spatial frequency images is greater than a preset frequency threshold, and to determine the degree of ablation of the water-blocking buffer layer of the cable according to the target image.

[0155] In this embodiment, visible light images can better display the physical damage and material changes of the surface, but have low sensitivity to temperature distribution and thermal damage. Infrared images can accurately capture temperature anomalies, but lack detailed description of surface texture and structural damage. Therefore, this solution combines the characteristics of these two images and fuses them together, thereby ensuring that the target image has the characteristics of both images. This provides a more comprehensive data foundation for subsequent ablation degree detection.

[0156] In the specific implementation process, the first determining unit includes an extraction module, a construction module, and a processing module. The extraction module extracts the features of the target image to obtain image features. The construction module is used to form a training set by combining historical image features and corresponding degree labels, and uses the training set to train an image recognition model to obtain a degree recognition model. The degree labels are the historical ablation degrees of the historical image features in the training set. The processing module is used to input the image features into the degree recognition model to obtain the ablation degree corresponding to the image features.

[0157] In this scheme, the deep learning-based ablation severity recognition model, trained on a large amount of historical data, can automatically learn the complex mapping relationship between image features and ablation severity, enabling accurate identification even for minor ablation with inconspicuous features or severe ablation with complex structures. The construction of the training set and the setting of labels ensure the model's generalization ability and prediction accuracy. The diversity of historical image features helps the model learn feature representations under different ablation patterns, while accurate ablation severity labels enable the model to accurately classify the ablation severity, thus ensuring accurate assessment of the ablation severity.

[0158] In some embodiments, the extraction module includes a first extraction submodule, a second extraction submodule, a control submodule, and a dimensionality reduction submodule. The first extraction submodule is used to extract at least one spatial frequency image corresponding to the target image whose spatial frequency is greater than the preset frequency threshold, and to determine the portion of the spatial frequency image whose spatial frequency is greater than the preset frequency threshold in the target image as a preliminary ablation region. The second extraction submodule is used to extract the center point or spatial frequency peak point of the preliminary ablation region in the target image to obtain seed points. The control submodule is used to control the growth of the seed points in the preliminary ablation region and to perform smoothing processing using anisotropic diffusion filtering technology to obtain the ablation region. The dimensionality reduction submodule is used to extract texture features from the ablation region and to perform dimensionality reduction using principal component analysis to obtain the image features.

[0159] In this scheme, multi-scale decomposition can decompose the image according to different frequencies, thereby capturing the ablation features at different scales. The selection of seed points is based on the center point or frequency peak point of the initial ablation region, which can guide the region growing algorithm to more accurately segment the ablation region, rather than treating the entire cable surface as the processing object. The selection of seed points and the application of the region growing algorithm realize the automatic expansion of the ablation region. Even under complex backgrounds and noise interference, a clear and complete ablation region image can be obtained. The smoothing processing of anisotropic diffusion filtering technology effectively suppresses image noise and maintains the sharpness of the ablation region boundary. The dimensionality reduction processing of principal component analysis identifies the main change directions in the original feature vector through statistical analysis. These directions are used as the feature axes after dimensionality reduction, which significantly reduces the dimensionality of the feature vector. At the same time, redundant information and noise interference are removed, thereby ensuring that the quality of the extracted image features is good.

[0160] In the specific implementation process, the above-mentioned device further includes a second acquisition unit and an addition unit. The second acquisition unit is used to acquire the image recognition model to be trained before training the image recognition model with the above-mentioned training set to obtain the degree recognition model, wherein the image recognition model to be trained is a residual network model. The addition unit is used to add an attention mechanism module between the convolutional layer and the activation function layer of the residual block of the image recognition model to be trained, so that the output of the convolutional layer is used as the input of the attention mechanism module, and the output of the attention mechanism module is used as the input of the activation function layer to obtain the image recognition model.

[0161] In this scheme, the residual network model maintains the training stability and performance of deep networks through residual connections. However, traditional residual network models treat all input features equally and lack the ability to focus on key features. Therefore, this scheme adds an attention mechanism module to each residual block, enabling the model to automatically learn and adjust the weights of different channels or features, allocating more computational resources to the regions where ablation features are most obvious, thereby improving the accuracy of feature recognition.

[0162] In some embodiments, the first fusion unit includes a first preprocessing module, a second preprocessing module, an acquisition module, and a fusion module. The first preprocessing module is used to preprocess the visible light image to obtain a preprocessed visible light image, wherein the preprocessing method includes filtering and / or image enhancement. The second preprocessing module is used to preprocess the infrared image to obtain a preprocessed infrared image, wherein the preprocessing method includes filtering and / or image enhancement. The acquisition module is used to acquire a first weighting coefficient and a second weighting coefficient, wherein the first weighting coefficient is the weighting coefficient of the preprocessed visible light image, and the second weighting coefficient is the weighting coefficient of the preprocessed infrared image. The fusion module is used to perform weighted fusion of the preprocessed visible light image and the preprocessed infrared image according to the first weighting coefficient and the second weighting coefficient to obtain the target image.

[0163] In this scheme, the preprocessing step removes noise from the image using filtering algorithms, improving the signal-to-noise ratio and making the features of the ablation area more prominent. Even with slight texture variations on the cable surface or inconsistent lighting conditions, the ablation area can be accurately detected. Image enhancement techniques, such as contrast stretching and histogram equalization, further enhance the contrast and clarity of the ablation area, making it more visible in the image and facilitating subsequent feature extraction and deep learning model recognition. The weighted fusion process dynamically adjusts the weights of the visible light and infrared images, ensuring that the ablation features are fully represented in the target image under different detection conditions. This avoids the limitations of information from a single image source and improves the comprehensiveness and accuracy of ablation degree detection.

[0164] In the specific implementation process, the above-mentioned device further includes a conversion unit, a first processing unit, a second processing unit, and a second fusion unit. The conversion unit is used to convert the target image from the time domain to the frequency domain using Fourier transform after determining the degree of ablation of the water-blocking buffer layer of the cable, thereby obtaining a frequency domain target image. The first processing unit is used to process the frequency domain target image using an unwrapping algorithm to obtain a depth frequency domain target image. The second processing unit is used to process the frequency domain target image using an edge enhancement algorithm to obtain a micro-frequency domain target image. The second fusion unit is used to perform weighted fusion of the depth frequency domain target image and the micro-frequency domain target image to obtain an ablation image, and then display the ablation image on the display interface.

[0165] In this scheme, Fourier transform extracts frequency component information from the target image, providing a foundation for subsequent depth information extraction and edge enhancement. The unwrapping algorithm processes the phase information obtained from the Fourier transform, generating a continuous depth map, avoiding discontinuities in depth information, and improving the accuracy of depth assessment. The edge enhancement algorithm enhances the edge information in the image, making the ablation edges and microstructures of the cable surface clearer, facilitating subsequent ablation feature spectrum extraction. By weighted fusion of the depth-frequency domain target image and the micro-frequency domain target image, a high-precision ablation image can be generated, which clearly displays the depth information and edge features of the ablation area.

[0166] In some embodiments, the above-described apparatus further includes a third acquisition unit, a calculation unit, a fourth acquisition unit, and a second determination unit. The third acquisition unit is used to acquire evaluation index information of the water-blocking buffer layer of the cable after determining the degree of ablation of the water-blocking buffer layer. The evaluation index information includes one or more of the following: degree of ablation, material strength, integrity of the water-blocking buffer layer, ablation rate, cable load, ambient temperature, and vibration stress. The calculation unit is used to calculate a weighted average of the multiple evaluation index information to obtain an evaluation score. The fourth acquisition unit is used to acquire multiple standardized ranges, wherein the standardized ranges are predefined score intervals of the cable's operational risk. The second determination unit is used to determine the standardized range in which the evaluation score falls as a target standardized range and determine the operational risk corresponding to the target standardized range as the current operational risk of the cable.

[0167] In this scheme, the multi-indicator comprehensive evaluation can fully reflect the health status of the cable's water-blocking buffer layer, avoiding the limitations of single-indicator evaluation. The calculation of the weighted average value takes into account the relative importance of different indicators in risk assessment, making the evaluation results more reasonable and reflecting the comprehensive level of cable operation risk.

[0168] The aforementioned cable water-blocking buffer layer erosion detection device includes a processor and a memory. The first acquisition unit, first fusion unit, decomposition unit, and first determination unit are all stored as program units in the memory. The processor executes these program units stored in the memory to achieve their respective functions. All of the above modules are located in the same processor; alternatively, the modules may be located in different processors in any combination.

[0169] The processor contains a kernel, which retrieves the corresponding program unit from memory. One or more kernels can be configured, and adjusting kernel parameters can address the problem in existing technologies where single-dimensional data cannot accurately assess the degree of ablation.

[0170] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0171] This invention provides a computer-readable storage medium, which includes a stored program, wherein when the program is executed, it controls the device containing the computer-readable storage medium to perform the cable water-blocking buffer layer ablation detection method.

[0172] This invention provides a processor for running a program, wherein the program executes the above-mentioned method for detecting the degree of erosion of the cable water-blocking buffer layer.

[0173] This invention provides a device including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements at least the steps of a method for detecting the degree of erosion of a cable's water-blocking buffer layer. The device described herein can be a server, PC, PAD, mobile phone, etc.

[0174] This application also provides a computer program product that, when executed on a data processing device, is adapted to perform a program that initializes a method for detecting the degree of erosion of a cable water-blocking buffer layer, including at least the following steps.

[0175] This application also provides a cable inspection system, including one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include methods for performing any of the above-described methods for detecting the degree of erosion of the cable's water-blocking buffer layer.

[0176] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0177] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0178] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0179] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0180] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0181] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0182] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0183] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0184] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0185] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0186] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for detecting the degree of erosion of the water-blocking buffer layer of a cable, characterized in that, include: Acquire visible light and infrared images of the water-blocking buffer layer of the cable; The visible light image and the infrared image are weighted and fused to obtain the target image; The target image is decomposed into multiple spatial frequency images corresponding to the target image, wherein the multi-scale decomposition is to decompose the image into different spatial frequencies; If the spatial frequency corresponding to at least one of the spatial frequency images is greater than a preset frequency threshold, it is determined that the water-blocking buffer layer of the cable is ablated, and the degree of ablation of the water-blocking buffer layer of the cable is determined according to the target image.

2. The method according to claim 1, characterized in that, Determining the degree of ablation of the water-blocking buffer layer of the cable based on the target image includes: Extract the features of the target image to obtain image features; A training set is formed by combining historical image features and corresponding degree labels. An image recognition model is trained using the training set to obtain a degree recognition model. The degree label is the historical ablation degree of the historical image features in the training set. The image features are input into the degree recognition model to obtain the ablation degree corresponding to the image features.

3. The method according to claim 2, characterized in that, Extracting features from the target image to obtain image features includes: Extract at least one spatial frequency image corresponding to the target image whose spatial frequency is greater than the preset frequency threshold, and determine the portion of the spatial frequency image whose spatial frequency is greater than the preset frequency threshold in the target image as the initial ablation area; Extract the center point or spatial frequency peak point of the preliminary ablation region in the target image to obtain seed points; The seed points are controlled to grow in the initial ablation region, and smoothed using anisotropic diffusion filtering technology to obtain the ablation region; Texture features are extracted from the ablation region and dimensionality reduction is performed using principal component analysis to obtain the image features.

4. The method according to claim 2, characterized in that, Before training the image recognition model using the training set to obtain the degree recognition model, the method further includes: Obtain an image recognition model to be trained, wherein the image recognition model to be trained is a residual network model; An attention mechanism module is added between the convolutional layer and the activation function layer of the residual block of the image recognition model to be trained, so that the output of the convolutional layer is used as the input of the attention mechanism module, and the output of the attention mechanism module is used as the input of the activation function layer to obtain the image recognition model.

5. The method according to claim 1, characterized in that, The visible light image and the infrared image are weighted and fused to obtain the target image, including: The visible light image is preprocessed to obtain a preprocessed visible light image, wherein the preprocessing method includes filtering and / or image enhancement; The infrared image is preprocessed to obtain the preprocessed infrared image, wherein the preprocessing method includes filtering and / or image enhancement; Obtain a first weighting coefficient and a second weighting coefficient, wherein the first weighting coefficient is the weighting coefficient of the preprocessed visible light image, and the second weighting coefficient is the preprocessed infrared image; The preprocessed visible light image and the preprocessed infrared image are weighted and fused according to the first weight coefficient and the second weight coefficient to obtain the target image.

6. The method according to any one of claims 1 to 5, characterized in that, After determining the degree of erosion of the water-blocking buffer layer of the cable, the method further includes: The target image is converted from the time domain to the frequency domain using Fourier transform to obtain the frequency domain target image; The frequency domain target image is processed using an unwrapping algorithm to obtain a depth frequency domain target image; An edge enhancement algorithm is used to process the frequency domain target image to obtain a micro-frequency domain target image; The ablation image is obtained by weighted fusion of the depth frequency domain target image and the micro frequency domain target image, and then displayed on the display interface.

7. The method according to any one of claims 1 to 5, characterized in that, After determining the degree of erosion of the water-blocking buffer layer of the cable, the method further includes: Obtain evaluation index information of the water-blocking buffer layer of the cable, wherein the evaluation index information includes one or more of the following: ablation degree, material strength, integrity of the water-blocking buffer layer, ablation rate, cable load, ambient temperature, and vibration stress. Calculate the weighted average of the multiple evaluation indicators to obtain the evaluation score; Obtain multiple standardized ranges, wherein the standardized ranges are predefined score intervals of the operational risk of the cable; The standardized range in which the evaluation score falls is determined as the target standardized range, and the operational risk corresponding to the target standardized range is determined as the current operational risk of the cable.

8. A device for detecting the degree of erosion of a cable's water-blocking buffer layer, characterized in that, include: The first acquisition unit is used to acquire visible light and infrared images of the water-blocking buffer layer of the cable; The first fusion unit is used to perform weighted fusion of the visible light image and the infrared image to obtain the target image; The decomposition unit is used to perform multi-scale decomposition on the target image to obtain multiple spatial frequency images corresponding to the target image, wherein the multi-scale decomposition is to decompose the image into different spatial frequencies; The first determining unit is configured to determine the ablation of the water-blocking buffer layer of the cable and the degree of ablation of the water-blocking buffer layer of the cable when the spatial frequency corresponding to at least one of the spatial frequency images is greater than a preset frequency threshold.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the cable water-blocking buffer layer ablation detection method according to any one of claims 1 to 7.

10. A cable testing system, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including methods for performing the cable water-blocking buffer layer ablation detection method according to any one of claims 1 to 7.

Images