Cable bridge fault detection method and system based on deep learning

Abstract

The invention relates to a cable bridge fault detection method and system based on deep learning, and the method comprises the following steps: carrying out the multispectral image collection of a region of a cable bridge, and obtaining a multispectral image data set; performing background modeling and noise suppression on the multispectral image data set through a multiscale Gaussian mixture model to obtain a multispectral enhanced image data set; performing cable bridge edge-gradient double-constraint segmentation on the multispectral enhanced image data set to obtain a bridge structure contour feature map; performing anomaly detection on the bridge structure contour feature map to obtain a bridge structure integrity anomaly map; and performing deformation-corrosion joint identification analysis on the cable bridge based on the bridge structural integrity anomaly graph to obtain a bridge fault type marking graph, thereby solving the technical problems of low efficiency, high cost, large subjective judgment error and difficulty in discovering a hidden fault in a traditional mode depending on manual inspection.

Description

Technical Field

[0001] The present invention relates to the technical field of cable trays, and in particular to a cable tray fault detection method and system based on deep learning. Background Art

[0002] The safe and stable operation of the power system is crucial to economic and social development. As key facilities that carry cables and ensure power transmission, the operating conditions of cable trays directly affect the reliability of power supply. As the scale of the power network continues to expand, cable trays are distributed over a wider area and their number has increased dramatically. They are exposed to complex environments for a long time and are susceptible to mechanical stress, chemical corrosion, natural aging, and other factors, leading to deformation, corrosion, and other faults. However, the traditional reliance on manual inspections is not only inefficient and costly, but also suffers from large subjective judgment errors and difficulty in detecting hidden faults. This makes it impossible to meet the needs of modern power systems for efficient cable tray operation and maintenance. Existing cable tray fault detection technologies, based on single spectral images, have limitations. Because different spectral bands reflect cable tray material properties and damage characteristics differently, a single spectrum cannot fully and accurately capture fault information, leading to frequent missed and false detections. Regarding image processing, conventional algorithms struggle to effectively extract cable tray structural features when faced with complex backgrounds and noise interference. The segmentation accuracy of cable tray edges and contours is insufficient, impacting the accuracy and reliability of subsequent fault identification and making them difficult to adapt to diverse on-site inspection environments. Deep learning, with its powerful feature extraction and pattern recognition capabilities, has shown great potential in the field of image detection and is gradually being applied to cable tray fault detection research. However, current deep learning-based detection methods are still imperfect and lack effective utilization of multispectral data. Key aspects such as background modeling, image segmentation, and joint identification of multiple fault types suffer from poor algorithm robustness and weak generalization capabilities. This makes it impossible to achieve high-precision, intelligent detection of cable tray faults. There is an urgent need to explore more efficient and reliable deep learning detection methods. Summary of the Invention

[0003] The main purpose of this invention is to provide a cable tray fault detection method based on deep learning, which solves the technical problems of the traditional method that relies on manual inspection, which is not only inefficient and costly, but also has large subjective judgment errors and difficulty in discovering hidden faults.

[0004] To achieve the above objectives, the present invention provides a cable tray fault detection method based on deep learning, comprising the following steps: Multispectral images of the cable bridge area are collected to obtain a multispectral image dataset; background modeling and noise suppression are performed on the multispectral image dataset through a multi-scale Gaussian mixture model to obtain a multispectral enhanced image dataset; cable bridge edge-gradient dual-constraint segmentation is performed on the multispectral enhanced image dataset to obtain a bridge structure contour feature map; anomaly detection is performed on the bridge structure contour feature map to obtain a bridge structure integrity anomaly map; based on the bridge structure integrity anomaly map, a deformation-corrosion joint identification analysis is performed on the cable bridge to obtain a bridge fault type annotation map.

[0005] Furthermore, the multispectral image dataset is subjected to background modeling and noise suppression by a multiscale Gaussian mixture model to obtain a multispectral enhanced image dataset, including: A multi-scale spatial pyramid is constructed for each band image in the multispectral image dataset through a multi-scale Gaussian mixture model to obtain a multi-scale image hierarchical structure; probability density estimation is performed on each level in the multi-scale image hierarchical structure to obtain a background-foreground separation confidence map, and background component modeling is performed on each pixel point in the multispectral image dataset based on the background-foreground separation confidence map to obtain a dynamic background reference field; non-uniform illumination compensation and multi-channel joint filtering processing are performed on the multispectral image dataset based on the dynamic background reference field to obtain a multispectral enhanced image dataset.

[0006] Furthermore, the cable tray edge-gradient dual-constraint segmentation is performed on the multispectral enhanced image dataset to obtain a cable tray structure contour feature map, including: Calculating the gradient intensity of the multispectral enhanced image data set by a multi-directional edge response operator to obtain an edge intensity distribution map, and performing directional angle quantization processing on the edge intensity distribution map to obtain an edge direction feature map; Performing adaptive cable tray threshold segmentation processing on the edge direction feature map to obtain an initial contour candidate area map, and extracting edge connectivity features of the initial contour candidate area map to obtain an edge connectivity domain feature map; Extracting an edge skeleton graph of the edge connected domain feature graph through morphological thinning operation, and performing chain coding conversion on the edge skeleton graph to obtain a bridge edge chain code sequence; Performing segmented fitting processing on the bridge frame contour based on the bridge frame edge chain code sequence to obtain a contour segment feature set, and extracting geometric features of the contour segment feature set to obtain a structural boundary feature vector; Performing spatial constraint verification on the structural boundary feature vector using a preset contour topology relationship analysis method to obtain an initial contour feature map; Based on the initial contour feature map, geometric structure optimization processing is performed to obtain a bridge structure contour feature map, wherein the bridge structure contour feature map includes the spatial geometric relationship between the bridge body boundary line, the supporting beam edge line and the connector contour line.

[0007] Furthermore, the edge skeleton graph is subjected to chain coding conversion to obtain a bridge edge chain code sequence, including: Performing eight-neighborhood scanning analysis on the edge skeleton image to obtain an edge pixel point direction sequence, and performing continuity start detection on the edge pixel point direction sequence to obtain an edge tracking starting point set; performing point-by-point tracking processing on the edge skeleton graph based on the edge tracking starting point set to obtain a point-by-point tracking edge point direction code table, and performing a differential operation on the point-by-point tracking edge point direction code table to obtain a direction change sequence, wherein the direction change sequence includes direction change amounts, cumulative angle values, and turning point marks between adjacent edge points; Performing run-length encoding compression on the direction change sequence to obtain an edge contour compression sequence, and marking edge segments of the edge contour compression sequence to obtain a contour segment description table, wherein the contour segment description table includes start and end positions of straight line segments and arc segments and descriptions of their geometric parameters; The edge skeleton graph is subjected to chain code reconstruction processing based on the contour segment description table to obtain a bridge edge chain code sequence.

[0008] Furthermore, the step of performing run-length coding compression on the direction change sequence to obtain an edge contour compression sequence includes: Performing continuity analysis on the direction change sequence to obtain continuous direction change intervals, and performing quantization processing on the continuous direction change intervals to obtain a quantized direction change table; Performing significance detection on the quantization direction change table by using an adaptive threshold segmentation technique to obtain a significant change paragraph set, and performing local feature extraction on the significant change paragraph set to obtain a local feature descriptor set; Performing run length statistics on the direction change sequence based on the local feature descriptor set to obtain a run length distribution graph, and performing entropy calculation on the run length distribution graph to obtain an entropy value sequence; The entropy value sequence is subjected to Huffman coding conversion to obtain a coding symbol sequence, and the direction change sequence is subjected to variable-length compression mapping based on the coding symbol sequence to obtain an edge contour compression sequence.

[0009] Furthermore, the abnormality detection is performed on the bridge structure profile characteristic map to obtain the bridge structure integrity abnormality map, including: Performing multi-scale morphological analysis on the bridge structure contour feature map to obtain a structural deformation feature set, and performing geometric parameter quantization processing on the structural deformation feature set to obtain a bridge geometric anomaly description table; Constructing a multi-scale anomaly feature map through the description of the bridge geometric anomaly; Integrity verification of the topological structure of the multi-scale anomaly feature map to obtain a set of candidate structural defect regions, and spatial cluster analysis of the candidate structural defect regions to obtain a defect region feature vector; The defect area feature vector is spatially mapped using morphological reconstruction technology to obtain a structural anomaly marker map, and the structural anomaly marker map is subjected to integrity assessment analysis to obtain a bridge structure integrity anomaly map, wherein the bridge structure integrity anomaly map includes the damage level, defect distribution and risk assessment indicators of the bridge structure.

[0010] Furthermore, the deformation-corrosion joint identification analysis of the cable tray is performed based on the abnormal diagram of the cable tray structure integrity to obtain a cable tray fault type annotation diagram, including: Extracting local deformation features from the abnormal diagram of the bridge structure integrity to obtain structural deformation features, and performing curvature-stress joint analysis on the structural deformation features to obtain a deformation stress distribution diagram; Performing surface texture analysis on the cable tray based on the deformation stress distribution map to obtain a surface corrosion feature map, and performing multi-band spectral response fusion processing on the surface corrosion feature map to obtain a corrosion degree quantification map; Performing structural damage assessment on the corrosion degree quantification map through spatial correlation analysis to obtain a damage state characteristic map, and performing connector integrity verification on the damage state characteristic map to obtain a connection state assessment map; Multimodal feature fusion processing is performed based on the connection status assessment diagram to obtain a preliminary fault type judgment diagram, and fault annotation is performed on the preliminary fault type judgment diagram to obtain a bridge fault type annotation diagram, wherein the bridge fault type annotation diagram includes classification labels and spatial coverage areas of typical fault types such as bending deformation, crack extension, rust plaque, and loose connection.

[0011] The present invention also provides a cable tray fault detection system based on deep learning, comprising: An acquisition module is used to acquire multispectral images of the area of ​​the cable bridge to obtain a multispectral image dataset; a suppression module is used to perform background modeling and noise suppression on the multispectral image dataset through a multi-scale Gaussian mixture model to obtain a multispectral enhanced image dataset; a segmentation module is used to perform cable bridge edge-gradient dual-constraint segmentation on the multispectral enhanced image dataset to obtain a bridge structure contour feature map; a detection module is used to perform anomaly detection on the bridge structure contour feature map to obtain a bridge structure integrity anomaly map; an identification module is used to perform deformation-corrosion joint identification analysis on the cable bridge based on the bridge structure integrity anomaly map to obtain a bridge fault type annotation map.

[0012] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0013] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0014] The present invention provides a cable tray fault detection method based on deep learning, comprising the following steps: acquiring multispectral images of the cable tray area to obtain a multispectral image dataset; performing background modeling and noise suppression on the multispectral image dataset using a multi-scale Gaussian mixture model to obtain a multispectral enhanced image dataset; performing cable tray edge-gradient dual-constraint segmentation on the multispectral enhanced image dataset to obtain a bridge structure contour feature map; performing anomaly detection on the bridge structure contour feature map to obtain a bridge structure integrity anomaly map; and performing deformation-corrosion joint identification analysis on the cable tray based on the bridge structure integrity anomaly map to obtain a bridge fault type annotation map. This method solves the technical problems of traditional manual inspection methods, which are not only inefficient and costly, but also suffer from large subjective judgment errors and difficulty in detecting hidden faults. By implementing anomaly detection on the bridge structure contour feature map, abnormal parts in the bridge structure can be quickly identified to obtain a structural integrity anomaly map. This process can promptly detect potential fault hazards, help operation and maintenance personnel quickly locate problem areas, and improve the efficiency of troubleshooting. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 1 is a schematic diagram of the steps of a cable tray fault detection method based on deep learning in one embodiment of the present invention; Figure 2 This is a block diagram of a cable tray fault detection system based on deep learning in one embodiment of the present invention; Figure 3It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0016] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0018] like Figure 1 As shown, Figure 1 This is a schematic diagram of the steps of a cable tray fault detection method based on deep learning in one embodiment of the present invention; In one embodiment of the present invention, a cable tray fault detection method based on deep learning is provided, comprising the following steps: Step S1: collecting multispectral images of the cable tray area to obtain a multispectral image dataset.

[0019] Specifically, in the power system, cable trays are important carriers of power transmission, and their operating status is related to the stability of the power supply. In order to comprehensively and accurately obtain the status information of cable trays, multispectral image acquisition of the cable tray area is adopted to obtain a multispectral image dataset. Multispectral image acquisition is to use imaging equipment that can sense different spectral bands to photograph the cable tray area. Different spectral bands can reflect different information such as the material properties, surface state, and potential damage of the bridge. For example, in the near-infrared spectral band, material changes caused by corrosion of the bridge can be more clearly detected; in the visible light band, cracks, deformations and other appearance characteristics of the bridge surface can be intuitively obtained. By simultaneously acquiring images in multiple spectral bands, these images with different spectral information are integrated to construct a multispectral image dataset. Taking the cable tray inspection in large substations as an example, the layout of the cable trays in the substation is complex, and the lighting conditions in some areas are different. Through multispectral image acquisition, the status of the cable tray in a complex environment can be captured from different spectral angles, and multi-band images such as visible light and infrared light can be fused into a data set, laying a data foundation for subsequent more accurate analysis of the cable tray condition.

[0020] Step S2: performing background modeling and noise suppression on the multispectral image dataset using a multi-scale Gaussian mixture model to obtain a multispectral enhanced image dataset.

[0021] Specifically, after acquiring a multispectral image dataset, due to the complex environment in which the cable bridge is located, such as the presence of various electrical equipment, pipelines, and other background interference in the substation, and the easy mixing of noise during the image acquisition process, which will affect the subsequent accurate analysis of the bridge status, it is necessary to use a multi-scale Gaussian mixture model to perform background modeling and noise suppression on the multispectral image dataset to obtain a multispectral enhanced image dataset. The multi-scale Gaussian mixture model can effectively adapt to changes in complex backgrounds by analyzing image data at different scales. In the model, the Gaussian distribution at each scale represents background areas and features of different scales in the image. By combining and iteratively calculating these Gaussian distributions, an accurate background model can be established. In terms of noise suppression, the model can identify pixels that deviate from the background model as noise, and then suppress them. For example, in a substation scenario, there are a large number of interference factors around the cable tray, such as reflections from metal equipment and shadows from pipelines. The multi-scale Gaussian mixture model can capture the background features of the cable tray's subtle structure at a small scale and process large-area complex backgrounds at a large scale. By continuously adjusting and optimizing the Gaussian distribution parameters, the background information can be accurately modeled while removing noise caused by light changes, sensor errors, etc., thereby enhancing the quality of the cable tray image and making the features of the cable tray more prominent in the multispectral image dataset, providing a clearer and more accurate data basis for subsequent image segmentation and fault detection.

[0022] Step S3: performing cable bridge edge-gradient dual-constraint segmentation on the multispectral enhanced image dataset to obtain a cable bridge structure contour feature map.

[0023] Specifically, after obtaining the multispectral enhanced image dataset, in order to accurately extract the structural information of the cable bridge, it is necessary to perform cable bridge edge-gradient dual-constraint segmentation on it, and then obtain the bridge structure contour feature map. Cable bridge edge-gradient dual-constraint segmentation uses the mutation characteristics of pixel grayscale values ​​at the bridge edge in the image and the change rules of gradient direction and amplitude to achieve accurate segmentation with dual constraints. First, by calculating the gradient amplitude of the image, it is possible to highlight the areas in the image where the grayscale changes drastically, that is, the possible locations of the bridge edge; at the same time, the gradient direction is analyzed so that the segmentation process is carried out along the direction of the bridge edge. In the actual application scenario of the substation, there are various complex backgrounds such as various equipment and lines around the cable bridge, and the lighting conditions may be uneven. Relying solely on a single edge detection or gradient calculation is prone to interference, resulting in segmentation errors. The edge-gradient dual-constraint segmentation method is used. For example, in dark corners, gradient calculation can enhance weak edge signals, and edge constraints can avoid misjudging the edges of other devices as cable tray edges. The two work together to effectively eliminate interference from complex backgrounds, accurately outline the contours of the cable tray, separate the cable tray from the background, and ultimately form a clear feature map of the cable tray structure, providing an accurate image basis for subsequent judgment of the integrity of the cable tray structure and identification of fault types.

[0024] Step S4: performing anomaly detection on the bridge structure profile characteristic map to obtain a bridge structure integrity anomaly map.

[0025] Specifically, after obtaining the cable tray structure contour feature map, since cable trays may suffer structural damage due to external forces, environmental erosion, and other factors during long-term operation, anomaly detection is required to obtain an anomaly map of the cable tray structure integrity. In the specific implementation process, a deep learning algorithm can be used to train a model using a large amount of normal cable tray structure contour feature map data, allowing the model to learn the characteristic patterns and distribution patterns of the normal cable tray structure. When the actual obtained cable tray structure contour feature map is input into the trained model, the model calculates the degree of difference between the input image and the normal pattern based on the learned normal pattern. If the difference exceeds a preset threshold, the area is determined to be abnormal. For example, in a substation scenario, cable trays are crisscrossed, and some cable trays may be deformed due to improper installation or long-term bearing of cable weight. By analyzing the contour feature map of the bridge structure through the trained anomaly detection model, abnormal areas such as bent bridge beams and deformed supports can be accurately identified, and these abnormal areas can be marked on the image to generate a bridge structure integrity anomaly map, which intuitively shows the damage to the bridge structure and provides an important basis for subsequent timely maintenance measures to ensure the stable operation of the power system.

[0026] Step S5: performing deformation-corrosion joint identification analysis on the cable tray based on the cable tray structural integrity anomaly map to obtain a cable tray fault type annotation map.

[0027] Specifically, after obtaining a cable tray structural integrity anomaly map, a combined deformation and corrosion identification analysis is performed on the cable tray to identify the specific fault type. This analysis then produces a labeled map of the cable tray fault types. This process is achieved by leveraging a deep learning classification model. Using a large number of cable tray image samples containing different fault types (such as deformation and corrosion), the model is trained to learn the characteristic patterns of different fault types in the images. The cable tray structural integrity anomaly map is then fed into the trained model, which then comprehensively analyzes various features of the anomaly area, such as texture, shape, and color. By calculating the degree of match between these features and various fault patterns, the model determines whether the fault is caused by deformation, corrosion, or a combination of these. For example, within a substation, cable trays exposed to humidity for extended periods of time may experience corrosion in some areas due to metal oxidation, coupled with deformation due to uneven load-bearing. Through this joint identification and analysis method, it is possible to accurately distinguish surface rust and peeling areas caused by corrosion on the cable tray, as well as bent and twisted parts caused by deformation, and mark the fault types with different symbols on the image to generate a cable tray fault type annotation map, which makes it convenient for operation and maintenance personnel to quickly understand the cable tray fault situation, thereby formulating scientific and reasonable maintenance plans to ensure the safe and stable operation of the cable tray.

[0028] In a specific embodiment, performing background modeling and noise suppression on the multispectral image dataset using a multiscale Gaussian mixture model to obtain a multispectral enhanced image dataset includes: A multi-scale spatial pyramid is constructed for each band image in the multispectral image dataset through a multi-scale Gaussian mixture model to obtain a multi-scale image hierarchical structure; probability density estimation is performed on each level in the multi-scale image hierarchical structure to obtain a background-foreground separation confidence map, and background component modeling is performed on each pixel point in the multispectral image dataset based on the background-foreground separation confidence map to obtain a dynamic background reference field; non-uniform illumination compensation and multi-channel joint filtering processing are performed on the multispectral image dataset based on the dynamic background reference field to obtain a multispectral enhanced image dataset.

[0029] Specifically, in power systems, complex environments such as substations where cable trays are located can introduce numerous interferences to multispectral image datasets, including equipment shadows, background noise from intersecting pipelines, and the effects of uneven illumination. Therefore, using a multiscale Gaussian mixture model to perform background modeling and noise suppression on the multispectral image dataset, thereby generating a multispectral enhanced image dataset, is crucial. The specific implementation process is as follows: First, a multiscale spatial pyramid is constructed using the multiscale Gaussian mixture model for each band of the multispectral image dataset, resulting in a multiscale image hierarchy. This is because images at different scales can capture distinct features of the cable tray and its background. Small-scale images can reveal details of the cable tray, while large-scale images can reflect the overall background layout. For example, when constructing a multiscale spatial pyramid for visible light images, assume that the original image is layer 0. By continuously downsampling the image, images at layers 1, 2, and so on are generated. If the original image resolution is 1024×1024 pixels, the first layer of the image might be downsampled to 512×512 pixels, the second layer to 256×256 pixels, and so on, forming a pyramid-like hierarchical structure. This allows the images to contain information of varying granularity at different levels. Smaller, higher-level images can focus on local details of the bridge, such as screws and welds, while larger, lower-level images encompass the entire bridge area and the surrounding complex background. This allows for a comprehensive representation of image information from multiple perspectives, providing a rich data foundation for subsequent processing. Next, a probability density estimation is performed for each level in the multiscale image hierarchy to generate a background-foreground separation confidence map. Based on this background-foreground separation confidence map, the background component of each pixel in the multispectral image dataset is modeled to obtain a dynamic background reference field. During the probability density estimation, for each pixel in each image level, a multiscale Gaussian mixture model calculates the probability of the pixel belonging to the background or foreground based on the grayscale and spectral characteristics of its neighboring pixels. For example, in a substation scenario, for a pixel at a certain level in a near-infrared band image from a multispectral image dataset, if the majority of surrounding pixels belong to the background area, such as the substation walls or ground, the model assigns a higher probability to that pixel as belonging to the background. By performing this probability calculation for all pixels at each level, a background-foreground separation confidence map for the entire level is generated. Assume that in a confidence map at a certain level, a value between 0 and 1 represents the probability of a pixel belonging to the background. The closer the value is to 1, the greater the probability that the pixel belongs to the background. By fusing and analyzing the confidence maps at each level, the background component of each pixel in the entire image can be determined, thereby constructing a dynamic background reference field. This dynamic background reference field adjusts as the image content and environment change. For example, if equipment moves within the substation, causing the background to change, the dynamic background reference field is updated accordingly to accurately reflect the current background state.Finally, the multispectral image dataset is subjected to non-uniform illumination compensation and multi-channel joint filtering based on a dynamic background reference field, resulting in a multispectral enhanced image dataset. In substations, the illumination of cable tray areas is often uneven due to factors such as equipment distribution and window locations, which can affect the accurate representation of cable tray features in the image. The dynamic background reference field can be used to analyze the region and degree of uneven illumination in the image. For example, if the grayscale of pixels in the upper left corner of the image is generally low, significantly different from the normal grayscale value for that area in the dynamic background reference field, it can be determined that insufficient illumination exists in that area. In this case, non-uniform illumination compensation is performed on this area, appropriately increasing its grayscale value to match the overall illumination conditions. Furthermore, multi-channel joint filtering combines information from different bands in the multispectral image dataset to denoise and enhance the image. For example, for visible and near-infrared images, certain noise manifests differently in different bands. Joint filtering can leverage the strengths of both bands to remove noise while preserving the key features of the cable tray. Assume that before processing, the image contains salt and pepper noise caused by sensor errors. After multi-channel joint filtering, the number of noise points is greatly reduced, from an average of 50 noise points per 100×100 pixel area before processing to an average of 5 noise points, which significantly improves the image quality. Finally, a multispectral enhanced image dataset is obtained, which provides clear and accurate data support for subsequent edge segmentation and fault detection of cable trays.

[0030] In a specific embodiment, performing edge-gradient dual-constraint segmentation of the cable bridge on the multispectral enhanced image dataset to obtain a cable bridge structure contour feature map includes: Calculating the gradient intensity of the multispectral enhanced image data set by a multi-directional edge response operator to obtain an edge intensity distribution map, and performing directional angle quantization processing on the edge intensity distribution map to obtain an edge direction feature map; Performing adaptive cable tray threshold segmentation processing on the edge direction feature map to obtain an initial contour candidate area map, and extracting edge connectivity features of the initial contour candidate area map to obtain an edge connectivity domain feature map; Performing skeleton extraction on the edge connected domain feature map through morphological thinning operation to obtain an edge skeleton map, and performing chain coding conversion on the edge skeleton map to obtain a bridge edge chain code sequence; Performing segmented fitting processing on the bridge frame contour based on the bridge frame edge chain code sequence to obtain a contour segment feature set, and performing geometric feature extraction on the contour segment feature set to obtain a structural boundary feature vector; The spatial constraint verification of the structural boundary feature vector is performed by a preset contour topology relationship analysis method to obtain an initial contour feature map, and geometric structure optimization processing is performed based on the initial contour feature map to obtain a bridge structure contour feature map, wherein the bridge structure contour feature map includes the spatial geometric relationship between the boundary line of the bridge body, the edge line of the supporting beam and the contour line of the connecting part.

[0031] Specifically, in the complex environment of a substation, cable trays are surrounded by various obstructions, such as electrical equipment and pipelines. Furthermore, lighting conditions are complex and variable, making accurate extraction of cable tray structural outlines from multispectral enhanced image datasets extremely challenging. Performing dual-constrained edge-gradient segmentation of cable trays within multispectral enhanced image datasets to generate a structural outline feature map is crucial for accurate cable tray fault detection. The implementation process is a step-by-step, refined process. First, a multi-directional edge response operator is used to calculate the gradient intensity of the multispectral enhanced image dataset to generate an edge intensity distribution map. This edge intensity distribution map is then quantified by directional angles to produce an edge directional feature map. The multi-directional edge response operator acts like a "detector" with multiple angles, sensitively capturing areas of image grayscale variation, potentially indicating edge locations, from multiple directions. Within a substation, the cable tray's horizontal beams, vertical columns, and diagonal support structures exhibit edge features in different directions. For example, eight directional edge response operators are set, corresponding to angles such as 0, 45, and 90 degrees. When calculating a multispectral enhanced image dataset, for horizontal bridge beams, under the action of the 0-degree operator, the gradient intensity of the pixel grayscale changes at the beam edge significantly increases. For example, in an image, the gradient intensity at the beam edge reaches 200 (assuming the gradient intensity range is 0-255), while the gradient intensity of the surrounding background area is only around 30. This highlights the bridge edge and generates an edge intensity distribution map. Next, to clarify the specific directional information of the edge, the edge intensity distribution map is quantized by directional angles. 360 degrees is divided into eight intervals of 45 degrees each, so that each edge pixel in the image corresponds to a specific directional angle interval. This results in an edge directional feature map containing the grayscale gradient values ​​and directional angle information of the bridge structure's horizontal, vertical, and diagonal edges, providing rich edge cues for subsequent segmentation. Next, the edge directional feature map is subjected to adaptive cable tray threshold segmentation to obtain an initial contour candidate region map. Edge connectivity features are extracted from this initial contour candidate region map to obtain an edge connected domain feature map. Due to the uneven lighting and complex background inside the substation, the grayscale representation of the bridge frame edges in the image varies, making it difficult to accurately extract the bridge frame outline using fixed threshold segmentation. Adaptive threshold segmentation can automatically determine the appropriate threshold based on the characteristics of different image regions. For example, in areas with strong lighting, the threshold is set to 150, while in areas with darker lighting, the threshold is set to 100. This accurately separates the bridge frame edges from the background and produces an initial contour candidate region map. However, this initial contour candidate region map may contain interfering information such as isolated pixels and discontinuous edge fragments, so it is necessary to extract its edge connectivity features.By analyzing the connectivity between pixels, if a pixel is connected to other pixels within a certain range (assuming a distance of 5 pixels), they are considered a connected domain. Isolated pixels are removed if they have no connected pixels within a certain range. This process yields an edge connectivity feature map containing the spatial location information of the bridge's surface edge connectivity, local structural breakpoints, and edge intersections, making the bridge's edge structure clearer and more complete. Morphological thinning is then used to extract the skeleton of the edge connectivity feature map, generating an edge skeleton map. This edge skeleton map is then chain-coded to produce a bridge edge chain code sequence. The goal of the morphological thinning operation is to gradually refine coarse edges into single-pixel-wide skeleton lines while maintaining edge connectivity. For example, in a substation cable tray image, a 5-pixel-wide bridge edge is transformed into a single-pixel-wide skeleton line after morphological thinning, more accurately representing the actual location and shape of the bridge edge. After obtaining the edge skeleton graph, it undergoes chain coding conversion using an 8-directional encoding scheme. The direction from one pixel to the next is represented by a number from 0 to 7, with 0 representing the right direction, 1 representing the upper right direction, and so on. This method records the directional relationships between edge pixels, generating a bridge edge chain code sequence containing information describing the bridge edge's direction, curvature changes, and connectivity. This provides an accurate digital edge description for subsequent contour fitting. Next, the bridge contour is segmented and fitted based on the bridge edge chain code sequence to obtain a segmented feature set. Geometric features are then extracted from this segmented feature set to obtain a structural boundary feature vector. The bridge edge chain code sequence records detailed information about the bridge edge's direction and connectivity. Due to the complexity of the bridge contour, direct processing is difficult, so segmented fitting is performed. For example, the irregularly shaped cable tray contour can be divided into 10 different line segments or curve segments for fitting. During the fitting process, methods such as the least squares method are used to find the segment equation or curve equation that best represents each segment. Geometric features are extracted for each segment, and information such as the endpoint coordinates, slope, and curvature of the curve are calculated. This results in a structural boundary feature vector containing the bridge edge segment equation, corner point coordinates, and contour closure index, making the geometric features of the bridge contour more clear and quantitative. Finally, the structural boundary feature vector is spatially constrained and verified using a preset topological relationship analysis method to obtain an initial contour feature map. Based on this initial contour feature map, geometric structure optimization is performed to obtain a bridge structure contour feature map. The preset topological relationship analysis method mainly considers the spatial positional relationship between the various parts of the bridge, such as the vertical relationship between the supporting beams and the bridge body, and the connection relationship between the connectors and the bridge edge.In the substation cable tray image, if the contour of a segment does not conform to the actual topological relationship, for example, the angle deviation between the contour of the supporting beam and the contour of the bridge body exceeds the allowable range of 5 degrees, the segment is adjusted. Through continuous verification and adjustment, the initial contour feature map is obtained. On this basis, the initial contour feature map is geometrically optimized and the corner points of the contour are smoothed to make the contour more consistent with the actual bridge structure. Finally, the bridge structure contour feature map containing the spatial geometric relationship of the bridge body boundary line, the supporting beam edge line and the connector contour line is obtained, which provides an accurate structural contour basis for subsequent cable tray fault detection. It should be noted that the specific implementation process of the preset contour topological relationship analysis method is: Explanation of the meaning of the letters in the formula: : Output initial contour feature map; : The preset contour topology relationship analysis function is used to verify the spatial constraints of boundary features; : Structural boundary feature vector, describing the geometric boundary information of the bridge structure; : The total number of boundary feature points; :Based on Feature points is the center and the radius is Neighborhood constraint area; : No. The position vector of the structural boundary feature points; : A predefined spatial constraint domain, which indicates the geometric space range where the legal contour is located.

[0032] In a specific embodiment, performing chain coding conversion on the edge skeleton graph to obtain a bridge edge chain code sequence includes: Performing eight-neighborhood scanning analysis on the edge skeleton image to obtain an edge pixel point direction sequence, and performing continuity start detection on the edge pixel point direction sequence to obtain an edge tracking starting point set; performing point-by-point tracking processing on the edge skeleton graph based on the edge tracking starting point set to obtain a point-by-point tracking edge point direction code table, and performing a differential operation on the point-by-point tracking edge point direction code table to obtain a direction change sequence, wherein the direction change sequence includes direction change amounts, cumulative angle values, and turning point marks between adjacent edge points; Performing run-length encoding compression on the direction change sequence to obtain an edge contour compression sequence, and marking edge segments of the edge contour compression sequence to obtain a contour segment description table, wherein the contour segment description table includes start and end positions of straight line segments and arc segments and descriptions of their geometric parameters; The edge skeleton graph is subjected to chain code reconstruction processing based on the contour segment description table to obtain a bridge edge chain code sequence.

[0033] Specifically, in the complex power facility environment of a substation, while the edge skeleton diagram of a cable tray already represents the basic outline of the tray structure, to more accurately describe its edge features and facilitate subsequent fault analysis and detection, the edge skeleton diagram must be chain-encoded to obtain a tray edge chain code sequence. This process is accomplished through multiple closely related steps, which are explained in detail below using the actual scenario of a substation cable tray. First, an eight-neighborhood scanning analysis is performed on the edge skeleton diagram to obtain a sequence of edge pixel directions. This sequence is then tested for continuity, resulting in a set of edge tracking starting points. The eight-neighborhood scanning analysis involves analyzing the neighboring pixels in each of the eight directions (up, down, left, right, upper left, upper right, lower left, and lower right) for each pixel in the edge skeleton diagram. In the edge skeleton diagram of a substation cable tray, due to the complex structure of the tray, including beams, columns, and braces, the edge skeleton exhibits an irregular shape. For example, for a pixel located on the edge of a cable bridge beam, an eight-neighborhood scan is performed. If the pixel to the right also belongs to the edge skeleton, the direction from that pixel to the pixel to the right is recorded. This scan of the entire edge skeleton image yields a sequence of edge pixel directions that reflects the directional relationships between each pixel. This sequence is then subjected to continuity start detection to identify the starting points and key nodes of the edge skeleton. For example, in cable bridge structures, the endpoints of the skeleton (such as the ends of the beam) and bifurcation points (such as the connection between the beam and the column) are important starting points. For example, in an edge skeleton image containing multiple cable bridges, detection reveals 10 distinct endpoints and 5 bifurcation points. The coordinates of these points and their local directional information (such as the edge extension direction at the endpoints and the directions of the branches at the bifurcation points) constitute the edge tracking starting point set. This information provides the key starting basis for accurate tracking of the cable bridge edge. Next, the edge skeleton image is tracked point by point based on the edge tracking starting point set, generating a point-by-point tracking edge point direction code table. This table is then differentially computed to produce a directional change sequence. Starting from a point in the edge tracking starting point set, each pixel is tracked sequentially along the direction of the edge skeleton. During the tracking process, the direction of each pixel is encoded using an eight-direction coding rule (0-7 corresponding to eight directions) based on the directional relationship between adjacent pixels. For example, starting from a starting point, if the next pixel is to the upper right of the starting point, the direction code for that pixel is 1. Through point-by-point tracking, the direction codes of the pixels on the entire edge skeleton are recorded, forming a point-by-point tracking edge point direction code table. This table is then differentially computed, calculating the difference between the direction codes of adjacent pixels, to produce a directional change sequence.This sequence contains rich information, such as the directional change between adjacent edge points, which intuitively reflects the degree of edge turning; the cumulative angle value, which reflects the overall curvature of a segment; and the turning point markers, which clearly indicate locations where the edge direction changes significantly. In a practical application of substation cable trays, suppose a segment of the cable tray edge is coded as [0, 1, 1, 2, 2, 3] from point-by-point tracking. After a difference operation, the resulting directional change sequence shows the directional changes of adjacent points as [1, 0, 1, 0, 1]. This indicates that the edge segment undergoes multiple directional changes during tracking, and the cumulative angle value increases accordingly. Points with significant directional changes are turning points. This information allows for a more detailed description of the shape characteristics of the cable tray edge. The directional change sequence is then run-length encoded to produce a compressed edge contour sequence. The edge segments of the compressed edge contour sequence are then marked to form a contour segment description table. Run-length encoding is a data compression method that combines consecutively repeated codes to reduce the data volume. In a directional change sequence, there may be multiple consecutive identical directional changes or cumulative angle values. For example, if there are five consecutive segments with directional changes of zero, run-length encoding can be used to represent these segments as "0 (5 times)", significantly shortening the data length and resulting in a compressed edge contour sequence. Subsequently, the edges are segmented and labeled based on the directional changes and characteristics within the compressed edge contour sequence. The edges of a substation cable tray can be categorized into different types, such as straight segments and arc segments. For example, by analyzing the compressed sequence, a segment from the 10th to the 20th point is identified as a straight segment with clear start and end locations. Geometric parameters of this segment, such as slope and endpoint coordinates, can be further extracted. Another segment from the 21st to the 35th point is an arc segment, and its start and end locations can also be determined, with geometric parameters such as the arc radius and central angle calculated. This information forms a contour segment description table, allowing the structural features of the cable tray edge to be clearly presented in a segmented manner. Finally, the edge skeleton graph is chain-coded and reconstructed based on the contour segment description table to obtain a bridge edge chain code sequence. The contour segment description table records the characteristic information of each bridge edge segment in detail. Based on this information, the edge skeleton graph is recoded and reconstructed. The geometric parameters and directional characteristics of each segment are integrated into the chain code to form a complete and concise bridge edge chain code sequence. In the actual substation cable tray inspection scenario, the resulting bridge edge chain code sequence can accurately describe the shape, directional changes, segment structure and other characteristics of the entire bridge edge.For example, the chain code sequence can clearly tell whether the edge of a certain section of bridge is a straight line or an arc, and its specific parameters such as length and angle. This provides an accurate and easy-to-process data basis for subsequent bridge profile analysis and fault detection based on the chain code sequence, making it possible to more efficiently identify possible deformation, damage and other problems of the bridge, thereby ensuring the stable operation of the power system.

[0034] In a specific embodiment, performing run-length coding compression on the direction change sequence to obtain an edge contour compression sequence includes: Performing continuity analysis on the direction change sequence to obtain continuous direction change intervals, and performing quantization processing on the continuous direction change intervals to obtain a quantized direction change table; Performing significance detection on the quantization direction change table by using an adaptive threshold segmentation technique to obtain a significant change paragraph set, and performing local feature extraction on the significant change paragraph set to obtain a local feature descriptor set; Performing run length statistics on the direction change sequence based on the local feature descriptor set to obtain a run length distribution graph, and performing entropy calculation on the run length distribution graph to obtain an entropy value sequence; The entropy value sequence is subjected to Huffman coding conversion to obtain a coding symbol sequence, and the direction change sequence is subjected to variable-length compression mapping based on the coding symbol sequence to obtain an edge contour compression sequence.

[0035] Specifically, in the substation cable tray inspection scenario, while the direction change sequence contains rich information about the cable tray's edge features, the raw data is large and redundant, making it difficult to process and analyze. Therefore, run-length encoding compression of the direction change sequence to generate a compressed edge profile sequence is crucial for streamlining data and highlighting key features. The specific implementation closely revolves around the structural characteristics of the cable tray. First, continuity analysis is performed on the direction change sequence to obtain continuous direction change intervals, which are then quantized to produce a quantized direction change table. Continuity analysis aims to identify segments within the direction change sequence that are continuous and have similar change trends. In a direction change sequence for a substation cable tray edge, the direction change of a section of the bridge beam may be relatively stable, with multiple consecutive points experiencing small direction changes. For example, in a direction change sequence, from the 10th to the 30th point, the direction change between adjacent points consistently fluctuates between 0 and 1. These 20 points can be identified as a continuous direction change interval. After all such intervals are identified, quantization is performed. During quantization, quantization levels are assigned based on information such as the cumulative angle value of each interval. Assume that cumulative angle values ​​between 0 and 15 degrees are classified as level 1, 16 and 30 degrees as level 2, and so on. For the interval between the 10th and 30th points, if its cumulative angle value is 12 degrees, it is quantized to level 1. The start and end points of the interval (the 10th and 30th points), the cumulative angle value (12 degrees), and its corresponding quantization level (level 1) are recorded. This information is then used to form a quantized direction change table, which presents the characteristics of the direction change sequence in a more regular manner. Next, adaptive threshold segmentation is used to perform significance detection on the quantized direction change table, generating a set of significantly changed segments. Local features are then extracted from these segments to form a set of local feature descriptors. Adaptive threshold segmentation automatically determines an appropriate threshold based on the data distribution characteristics in the quantized direction change table, distinguishing segments with significant changes from relatively stable segments. In actual images of substation cable trays, edge direction changes are particularly pronounced at the junctions between different components or where deformation has occurred. For example, in the quantized directional change table, if the cumulative angle value of a certain interval changes significantly, exceeding an adaptive threshold (assuming it's automatically determined to be 20 degrees based on the data distribution), the paragraph containing that interval is identified as a significantly changed paragraph. In this way, all significantly changed paragraphs are identified, forming a significantly changed paragraph set. Local features are then extracted for each significantly changed paragraph, such as its position coordinates in the directional change sequence (the sequence numbers of the starting and ending points), the paragraph length (the number of points included), and curvature features (the approximate curvature calculated from the directional changes of adjacent points).Assume that a significant change segment begins at the 50th point and ends at the 70th point, with a length of 20 points. Its average curvature is calculated to be 0.3. This information constitutes a local feature descriptor set, highlighting the characteristics of key areas along the bridge edge. Then, based on this local feature descriptor set, run length statistics are performed on the directional change sequence to obtain a run length distribution graph, and entropy is calculated on this run length to obtain an entropy sequence. Run length statistics are calculated by counting the length of consecutive occurrences of the same data or similar features within the directional change sequence. Combining the local feature descriptor set, the run length is calculated for each significant change segment or stable segment. For example, in a directional change sequence corresponding to a stable straight edge segment, the same directional change may occur 15 times in a row, resulting in a run length of 15. This statistical analysis is performed on the entire directional change sequence to obtain a run length distribution graph, showing the distribution of different run lengths. The entropy value of each run is then calculated, reflecting the uncertainty or complexity of the data within that run. For example, a run with a run length of 10 and relatively uniform data may have a low entropy value; whereas a run with diverse data variations may have a higher entropy value. Through calculation, an entropy sequence containing information about each run length and its corresponding entropy value is obtained, further exploring the data characteristics of the direction change sequence. Finally, the entropy sequence is transformed by Huffman coding to obtain a code symbol sequence. Based on this code symbol sequence, variable-length compression mapping is performed on the direction change sequence to obtain an edge contour compression sequence. Huffman coding is an efficient data compression coding method that assigns codes of different lengths based on the frequency of occurrence of data, with codes of higher frequency resulting in shorter codes. In the entropy sequence, different entropy values ​​or run lengths are treated as different data types and Huffman coding is performed based on their frequency of occurrence to obtain a code symbol sequence. For example, if a run length occurs frequently, a shorter code is assigned; if it occurs less frequently, a longer code is assigned. Based on the code symbol sequence, variable-length compression mapping is performed on the original direction change sequence, replacing the data in the original sequence with the corresponding code symbol, significantly compressing the data. The resulting compressed edge profile sequence retains key information, such as the directional change and turning point markers of the bridge edge, while also reducing data storage and transmission costs through a compressed representation. For example, in the compressed sequence, the originally lengthy directional change data is reduced to compact coded symbols, while still accurately reflecting important features such as where the bridge edge turns and the degree of directional change. This provides efficient data support for subsequent bridge fault diagnosis and structural analysis based on the compressed sequence.

[0036] In a specific embodiment, the abnormality detection of the bridge structure profile characteristic map to obtain the bridge structure integrity abnormality map includes: Performing multi-scale morphological analysis on the bridge structure contour feature map to obtain a structural deformation feature set, and performing geometric parameter quantization processing on the structural deformation feature set to obtain a bridge geometric anomaly description table; Constructing a multi-scale anomaly feature map through the description of the bridge geometric anomaly; Integrity verification of the topological structure of the multi-scale anomaly feature map to obtain a set of candidate structural defect regions, and spatial cluster analysis of the candidate structural defect regions to obtain a defect region feature vector; The defect area feature vector is spatially mapped using morphological reconstruction technology to obtain a structural anomaly marker map, and the structural anomaly marker map is subjected to integrity assessment analysis to obtain a bridge structure integrity anomaly map, wherein the bridge structure integrity anomaly map includes the damage level, defect distribution and risk assessment indicators of the bridge structure.

[0037] Specifically, during the operation and maintenance of substation cable bridges, a multi-scale morphological analysis is performed on the bridge structure contour feature map to obtain a bridge structure integrity anomaly map. This is a step-by-step, interlocking process that can accurately identify potential structural defects in the bridge. First, a multi-scale morphological analysis is performed on the bridge structure contour feature map to obtain a structural deformation feature set, and the structural deformation feature set is quantified by geometric parameters to obtain a bridge geometric anomaly description table. Multi-scale morphological analysis is like observing the bridge structure with a magnifying glass of different magnifications, capturing various possible deformations from the macro to the micro. In actual operation, structural elements of different sizes are used to perform operations such as corrosion, expansion, opening and closing operations on the bridge structure contour feature map. For example, for a cable tray with a span of 5 meters, morphological operations using circular structural elements with a diameter of 50 cm during large-scale analysis can detect overall bending deformation of the cable tray, such as a 3 cm downward bend in the middle of the cable tray. At a smaller scale, circular structural elements with a diameter of 5 cm can identify subtle local deformations, such as a 10 cm long and 2 cm deep dent on a supporting beam. Through these multi-scale morphological operations, various deformation features of the cable tray structure are extracted, forming a structural deformation feature set. Next, geometric parameter quantification is performed on the structural deformation feature set. The extracted deformation features are converted into specific geometric parameters, such as length, angle, and area. For the aforementioned bending deformation in the middle of the cable tray, the angle is quantified to 0.35 degrees (calculated based on the bending height and span). For the dent on the supporting beam, the length is quantified to 10 cm, the width is 5 cm, and the area is 50 square centimeters. These quantified geometric parameters are organized into a table to create a description table of the cable tray's geometric anomalies. This table details the specific geometric parameters of each deformation feature, providing an accurate data foundation for subsequent analysis. A multi-scale anomaly feature map is constructed based on the bridge's geometric anomaly description. Using the data in the bridge's geometric anomaly description table, anomaly information at different scales is mapped into image space. During this process, different anomalies are assigned different colors, brightness, or texture features based on the size and nature of the geometric parameters. For example, large structural deformations are represented by red, while smaller local defects are represented by yellow. Furthermore, the brightness or saturation of the corresponding area is adjusted based on the severity of the anomaly. For example, suppose a severe bending deformation and multiple minor corrosion defects are detected in a cable tray. In the multi-scale anomaly feature map, the bending deformation area will appear bright red, while the corrosion defect area will appear lighter yellow. This visualization method allows for an intuitive understanding of the distribution of anomalies at different scales within the bridge structure. The topology of the multi-scale anomaly feature map is verified for integrity, resulting in a set of candidate structural defect regions. Spatial clustering analysis is then performed on this set of candidate structural defect regions to generate a defect region feature vector.Topological structure verification mainly checks the spatial connectivity and structural integrity between abnormal regions in the multi-scale anomaly feature map. In the image of the cable tray, the normal structure should have a continuous topological relationship, while the defective region often destroys this continuity. For example, when a crack appears in the cable tray, it will appear as a topological structure break in the multi-scale anomaly feature map. By analyzing these topological structural anomalies, areas with possible defects are identified, forming a set of candidate structural defect regions. Spatial clustering analysis is performed on the set of candidate structural defect regions. Adjacent defect regions with similar characteristics are grouped together based on the spatial location and feature similarity of the defect regions. For example, at the corner of a cable tray, multiple small corrosion areas were detected. Through spatial clustering analysis, these corrosion areas were clustered into a larger corrosion area. The characteristic parameters such as the center position, area, and shape of each clustered area are calculated to form a defect region feature vector. Suppose that cluster analysis identifies an area consisting of multiple small corrosion spots as an elliptical region with an area of ​​200 square centimeters, a center coordinate of (x=100, y=200), a major axis of 15 centimeters, and a minor axis of 10 centimeters. These parameters constitute the feature vector of the defect region. Morphological reconstruction techniques are used to spatially map the defect region feature vector to produce a structural anomaly marker map. This map is then subjected to an integrity assessment and analysis to produce a cable bridge structural integrity anomaly map. Morphological reconstruction techniques use the information in the defect region feature vectors to accurately identify the location and extent of the defect regions on the original image. Each defect region's feature vector, derived from the cluster analysis, is mapped onto the original image, and these regions are marked with specific colors or symbols to produce a structural anomaly marker map. For example, in the original cable bridge image, identified corrosion areas are marked with a red border, and cracked areas are marked with a blue line. The structural anomaly marker map is then subjected to an integrity assessment and analysis. The integrity of the cable bridge structure is assessed by comprehensively considering factors such as the size, number, location, and type of defect regions. A set of assessment criteria is developed to categorize bridge structure damage levels into different levels, such as minor, moderate, and severe, based on indicators such as the area percentage and severity of the defected areas. The distribution of defects is also analyzed to identify areas of higher risk. For example, in one cable tray, multiple areas of corrosion were discovered, totaling 5% of the tray's surface area, primarily concentrated in the supports. Based on the assessment criteria, the damage level of the cable tray was determined to be moderate, and the risk assessment indicated that the supports posed a high safety hazard. Ultimately, these assessment results are integrated to form a structural integrity anomaly map that includes the damage level, defect distribution, and risk assessment indicators.

[0038] In a specific embodiment, the deformation-corrosion joint identification analysis of the cable tray is performed based on the abnormal diagram of the cable tray structure integrity to obtain a cable tray fault type annotation diagram, including: Extracting local deformation features from the abnormal diagram of the bridge structure integrity to obtain structural deformation features, and performing curvature-stress joint analysis on the structural deformation features to obtain a deformation stress distribution diagram; Performing surface texture analysis on the cable tray based on the deformation stress distribution map to obtain a surface corrosion feature map, and performing multi-band spectral response fusion processing on the surface corrosion feature map to obtain a corrosion degree quantification map; Performing structural damage assessment on the corrosion degree quantification map through spatial correlation analysis to obtain a damage state characteristic map, and performing connector integrity verification on the damage state characteristic map to obtain a connection state assessment map; Multimodal feature fusion processing is performed based on the connection status assessment diagram to obtain a preliminary fault type judgment diagram, and fault annotation is performed on the preliminary fault type judgment diagram to obtain a bridge fault type annotation diagram, wherein the bridge fault type annotation diagram includes classification labels and spatial coverage areas of typical fault types such as bending deformation, crack extension, rust plaque, and loose connection.

[0039] Specifically, in the complex operating environment of power facilities in substations, cable trays are constantly subjected to multiple factors, including mechanical stress and environmental corrosion. It's difficult to comprehensively and accurately determine fault types based solely on the cable tray structural integrity anomaly map. Therefore, it's necessary to perform a combined deformation and corrosion analysis on the cable tray based on this map to generate a cable tray fault type annotation map, providing a detailed basis for operational and maintenance decisions. This process, through multi-dimensional, multi-step analysis, gradually delves into the fault characteristics of the cable tray. First, local deformation features are extracted from the cable tray structural integrity anomaly map to generate a structural deformation feature map. This map is then subjected to a combined curvature-stress analysis to produce a deformation-stress distribution map. Within a substation, cable trays can experience varying degrees of deformation due to factors such as the weight of cables and external forces. Local deformation feature extraction analyzes changes in the cable tray's structural contour to identify areas of deformation, such as bends and twists. For example, within the cable tray structural integrity anomaly map, an algorithm detects a significant bend in a 2-meter-long section of the cable tray beam. The features of this bend are then extracted to form a structural deformation feature map. Next, a combined curvature-stress analysis was performed on the structural deformation characteristic map. Curvature reflects the degree of bending of the bridge structure, while stress reflects the internal stress conditions. Assuming the maximum curvature of the curved beam section reaches 0.05 (unit: meter⁻¹), calculations and analysis indicate stress concentration at the most severe bending point, reaching a value of 150 MPa (assumed stress unit). This analysis then yields a deformation-stress distribution map, clearly demonstrating the degree of bending deformation, stress concentration areas, and the spatial distribution of structural strain, providing foundational data for subsequent analysis. Next, surface texture analysis of the cable tray was performed based on the deformation-stress distribution map to generate a surface corrosion characteristic map. This map was then subjected to multi-band spectral response fusion processing to quantify the degree of corrosion. Due to factors such as humidity and chemical gases present within the substation, the bridge surface is susceptible to corrosion. Surface texture analysis uses image processing techniques to identify corrosion texture features such as rust and flaking on the bridge surface. For example, if irregular rust patterns were observed on a particular bridge section, these texture features were extracted to form a surface corrosion characteristic map. Spectra in different wavelengths reflect the severity of corrosion differently. Multi-band spectral response fusion combines image information from multiple wavelengths, including visible and infrared light, to more accurately assess the extent of corrosion. For example, in the visible light band, a rusted area appears reddish-brown, but it's difficult to accurately determine the depth of corrosion. However, in the near-infrared band, changes in material thickness caused by corrosion are more clearly revealed. By fusing multi-band spectral information and conducting a quantitative analysis of the corroded area, the corrosion depth in this rusted area is determined to be 0.3 mm, with a material degradation level of 30% (assuming a quantitative indicator). This generates a quantitative corrosion map, which includes the rusted area on the bridge surface, the corrosion depth, and quantitative indicators of the material degradation level.Then, spatial correlation analysis is used to assess structural damage based on the corrosion quantification map, generating a damage state characteristic map. This map is then used to verify the integrity of connectors, resulting in a connection state assessment map. Spatial correlation analysis considers the spatial relationship between the corrosion area and the overall bridge structure to assess the impact of corrosion on structural performance. For example, corrosion located near critical load-bearing locations on the bridge, such as support connections, has a greater impact on structural damage. Through this analysis, the extent of structural damage from different corrosion areas is quantified, forming a damage state characteristic map. Connector integrity verification focuses on the connections between bridge components. In substations, connectors such as bolts and welds can become loose or damaged due to corrosion, vibration, and other factors. The connector area in the damage state characteristic map is inspected. For example, if a bolt is found to be loose by 20% (a hypothetical quantitative indicator), its stress state has changed. Calculation indicates that the structural stability parameter of this connection has decreased by 15%. This results in a connection state assessment map, which includes the looseness, stress state, and structural stability parameters of the bridge connection. Finally, multimodal feature fusion is performed on the connection status assessment image to generate a preliminary fault type map, which is then labeled to create a cable tray fault type annotation map. This multimodal feature fusion process integrates multiple aspects of information, including the deformation stress distribution map, corrosion quantification map, and connection status assessment map, to determine the fault type from multiple perspectives. For example, by combining information on cable tray bending deformation, corrosion, and loose connections, an algorithmic analysis determines that a section of cable tray suffers from multiple faults, including bending deformation, rust patches, and loose connections. This fault information is annotated on the image to form a preliminary fault type map. This preliminary fault type map is then further refined to clearly define the classification labels and spatial coverage areas for different fault types. For example, bending deformation areas are marked in red, labeled "bending deformation"; rust patches are marked in yellow, labeled "rust patches"; and loose connections are marked in blue, labeled "loose connections." This ultimately creates a clear and intuitive cable tray fault type annotation map, enabling operations and maintenance personnel to quickly locate the fault, understand the fault type, and implement targeted repair measures to ensure the safe and stable operation of the cable tray.

[0040] The above describes the cable tray fault detection method based on deep learning in the embodiment of the present invention. The following describes the cable tray fault detection system based on deep learning in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a cable tray fault detection system based on deep learning includes: An acquisition module 21 is used to acquire multispectral images of the area of ​​the cable bridge to obtain a multispectral image data set; a suppression module 22 is used to perform background modeling and noise suppression on the multispectral image data set through a multi-scale Gaussian mixture model to obtain a multispectral enhanced image data set; a segmentation module 23 is used to perform cable bridge edge-gradient dual-constraint segmentation on the multispectral enhanced image data set to obtain a bridge structure contour feature map; a detection module 24 is used to perform anomaly detection on the bridge structure contour feature map to obtain a bridge structure integrity anomaly map; an identification module 25 is used to perform deformation-corrosion joint identification analysis on the cable bridge based on the bridge structure integrity anomaly map to obtain a bridge fault type annotation map.

[0041] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0042] Reference Figure 3 The embodiment of the present invention further provides a computer device, the internal structure of which can be as follows Figure 3 As shown. The computer device includes a processor, memory, display screen, input device, network interface and database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0043] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0044] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the above-described method when executed by a processor. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0045] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware using a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media provided herein and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM.

[0046] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0047] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A cable tray fault detection method based on deep learning, characterized in that: The following steps are involved: Multispectral images of the cable bridge area are collected to obtain a multispectral image dataset; background modeling and noise suppression are performed on the multispectral image dataset using a multi-scale Gaussian mixture model to obtain a multispectral enhanced image dataset; and cable bridge edge-gradient dual-constraint segmentation is performed on the multispectral enhanced image dataset to obtain a bridge structure contour feature map; Anomaly detection is performed on the bridge structure contour feature map to obtain a bridge structure integrity anomaly map; based on the bridge structure integrity anomaly map, a deformation-corrosion joint identification analysis is performed on the cable bridge to obtain a bridge fault type annotation map.

2. The cable tray fault detection method based on deep learning according to claim 1 is characterized in that: The method of performing background modeling and noise suppression on the multispectral image dataset by using a multiscale Gaussian mixture model to obtain a multispectral enhanced image dataset includes: A multi-scale spatial pyramid is constructed for each band image in the multispectral image dataset through a multi-scale Gaussian mixture model to obtain a multi-scale image hierarchical structure; probability density estimation is performed on each level in the multi-scale image hierarchical structure to obtain a background-foreground separation confidence map, and background component modeling is performed on each pixel point in the multispectral image dataset based on the background-foreground separation confidence map to obtain a dynamic background reference field; non-uniform illumination compensation and multi-channel joint filtering processing are performed on the multispectral image dataset based on the dynamic background reference field to obtain a multispectral enhanced image dataset.

3. The cable tray fault detection method based on deep learning according to claim 1 is characterized in that: The cable bridge edge-gradient dual-constraint segmentation is performed on the multispectral enhanced image dataset to obtain a cable bridge structure contour feature map, including: Calculating the gradient intensity of the multispectral enhanced image data set by a multi-directional edge response operator to obtain an edge intensity distribution map, and performing directional angle quantization processing on the edge intensity distribution map to obtain an edge direction feature map; Performing adaptive cable tray threshold segmentation processing on the edge direction feature map to obtain an initial contour candidate area map, and extracting edge connectivity features of the initial contour candidate area map to obtain an edge connectivity domain feature map; Extracting an edge skeleton graph of the edge connected domain feature graph through morphological thinning operation, and performing chain coding conversion on the edge skeleton graph to obtain a bridge edge chain code sequence; Performing segmented fitting processing on the bridge frame contour based on the bridge frame edge chain code sequence to obtain a contour segment feature set, and extracting geometric features of the contour segment feature set to obtain a structural boundary feature vector; Performing spatial constraint verification on the structural boundary feature vector using a preset contour topology relationship analysis method to obtain an initial contour feature map; Based on the initial contour feature map, geometric structure optimization processing is performed to obtain a bridge structure contour feature map, wherein the bridge structure contour feature map includes the spatial geometric relationship between the bridge body boundary line, the supporting beam edge line and the connector contour line.

4. The cable tray fault detection method based on deep learning according to claim 3 is characterized in that: The chain coding conversion of the edge skeleton graph to obtain a bridge edge chain code sequence includes: Performing eight-neighborhood scanning analysis on the edge skeleton image to obtain an edge pixel point direction sequence, and performing continuity start detection on the edge pixel point direction sequence to obtain an edge tracking starting point set; performing point-by-point tracking processing on the edge skeleton graph based on the edge tracking starting point set to obtain a point-by-point tracking edge point direction code table, and performing a differential operation on the point-by-point tracking edge point direction code table to obtain a direction change sequence, wherein the direction change sequence includes direction change amounts, cumulative angle values, and turning point marks between adjacent edge points; Performing run-length encoding compression on the direction change sequence to obtain an edge contour compression sequence, and marking edge segments of the edge contour compression sequence to obtain a contour segment description table, wherein the contour segment description table includes start and end positions of straight line segments and arc segments and descriptions of their geometric parameters; The edge skeleton graph is subjected to chain code reconstruction processing based on the contour segment description table to obtain a bridge edge chain code sequence.

5. The cable tray fault detection method based on deep learning according to claim 4 is characterized in that: The step of performing run-length coding compression on the direction change sequence to obtain an edge contour compression sequence comprises: Performing continuity analysis on the direction change sequence to obtain continuous direction change intervals, and performing quantization processing on the continuous direction change intervals to obtain a quantized direction change table; Performing significance detection on the quantization direction change table by using an adaptive threshold segmentation technique to obtain a significant change paragraph set, and performing local feature extraction on the significant change paragraph set to obtain a local feature descriptor set; Performing run length statistics on the direction change sequence based on the local feature descriptor set to obtain a run length distribution graph, and performing entropy calculation on the run length distribution graph to obtain an entropy value sequence; The entropy value sequence is subjected to Huffman coding conversion to obtain a coding symbol sequence, and the direction change sequence is subjected to variable-length compression mapping based on the coding symbol sequence to obtain an edge contour compression sequence.

6. The cable tray fault detection method based on deep learning according to claim 1 is characterized in that: The abnormality detection is performed on the bridge structure contour feature map to obtain the bridge structure integrity abnormality map, including: Performing multi-scale morphological analysis on the bridge structure contour feature map to obtain a structural deformation feature set, and performing geometric parameter quantization processing on the structural deformation feature set to obtain a bridge geometric anomaly description table; Constructing a multi-scale anomaly feature map through the description of the bridge geometric anomaly; Integrity verification of the topological structure of the multi-scale anomaly feature map to obtain a set of candidate structural defect regions, and spatial cluster analysis of the candidate structural defect regions to obtain a defect region feature vector; The defect area feature vector is spatially mapped using morphological reconstruction technology to obtain a structural anomaly marker map, and the structural anomaly marker map is subjected to integrity assessment analysis to obtain a bridge structure integrity anomaly map, wherein the bridge structure integrity anomaly map includes the damage level, defect distribution and risk assessment indicators of the bridge structure.

7. The cable tray fault detection method based on deep learning according to claim 1 is characterized in that: The deformation-corrosion joint identification analysis of the cable tray is performed based on the abnormal diagram of the cable tray structure integrity to obtain a cable tray fault type annotation diagram, including: Extracting local deformation features from the abnormal diagram of the bridge structure integrity to obtain structural deformation features, and performing curvature-stress joint analysis on the structural deformation features to obtain a deformation stress distribution diagram; Performing surface texture analysis on the cable tray based on the deformation stress distribution map to obtain a surface corrosion feature map, and performing multi-band spectral response fusion processing on the surface corrosion feature map to obtain a corrosion degree quantification map; Performing structural damage assessment on the corrosion degree quantification map through spatial correlation analysis to obtain a damage state characteristic map, and performing connector integrity verification on the damage state characteristic map to obtain a connection state assessment map; Multimodal feature fusion processing is performed based on the connection status assessment map to obtain a preliminary fault type judgment map, and fault annotation is performed on the preliminary fault type judgment map to obtain a bridge fault type annotation map, wherein the bridge fault type annotation map includes classification labels and spatial coverage areas of typical fault types such as bending deformation, crack extension, rust plaque, and loose connection.

8. A cable tray fault detection system based on deep learning, characterized in that: include: an acquisition module, configured to acquire multispectral images of the area of ​​the cable tray to obtain a multispectral image dataset; a suppression module, configured to perform background modeling and noise suppression on the multispectral image dataset using a multi-scale Gaussian mixture model to obtain a multispectral enhanced image dataset; a segmentation module, configured to perform edge-gradient dual-constraint segmentation of the cable bridge on the multispectral enhanced image dataset to obtain a cable bridge structure contour feature map; The detection module is used to perform anomaly detection on the bridge structure contour feature map to obtain a bridge structure integrity anomaly map; the identification module is used to perform deformation-corrosion joint identification analysis on the cable bridge based on the bridge structure integrity anomaly map to obtain a bridge fault type annotation map.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images