Cable intermediate joint defect detection method and system based on X-ray imaging and deep learning

Abstract

The invention relates to the technical field of power equipment detection, and discloses a cable intermediate joint defect detection method and system based on X-ray imaging and deep learning, and the method comprises the steps: collecting a multi-energy and multi-angle X-ray image through preset scanning parameters; carrying out scattering correction and fusion processing on the acquired X-ray image; reconstructing a three-dimensional model of the cable intermediate joint based on the multi-angle projection data; and extracting defect features by using a multi-scale convolutional neural network and outputting a detection result. According to the method, the cable intermediate joint is detected by adopting an X-ray penetrating imaging mode at the cable intermediate joint, and the defects and defect hidden dangers of the cable intermediate joint can be effectively detected through methods of image processing, defect identification and the like.

Description

Technical Field

[0001] This invention relates to the field of power equipment testing technology, and in particular to a method and system for detecting defects in cable joints based on X-ray imaging and deep learning. Background Technology

[0002] With the rapid development of power systems, cable joints, as an important component of power transmission, are directly related to the stable operation of power systems in terms of quality and safety. However, defects that may occur in cable joints during manufacturing and use, such as cracks, voids, and foreign objects, pose a great threat to the safe operation of power systems. Therefore, how to efficiently and accurately detect defects in cable joints and ensure the safe operation of power systems has become an important issue that urgently needs to be addressed.

[0003] Existing technologies for cable joint inspection suffer from several shortcomings, including low detection accuracy, difficulty in identifying minute defects, limited detection methods that fail to comprehensively cover all defect types, complex and time-consuming processes, low efficiency, susceptibility to interference, and poor stability. This invention, by employing dual-energy X-ray data acquisition, multimodal image fusion, and intelligent defect diagnosis, effectively improves the accuracy and coverage of cable joint inspection, as well as enhancing detection efficiency and stability. As a crucial power transmission device, the quality inspection and installation of high-voltage cables are paramount. Cable terminations and intermediate joints are relatively weak points in the insulation structure of cable lines, making them prone to failure. Based on past cable operation data, approximately 10% of cable accidents are due to potential insulation defects caused by improper cable installation. Once a cable fault occurs, the repair and replacement time is significantly longer than for overhead lines, resulting in prolonged power outages at key cable terminals. Therefore, fault detection and diagnosis of operating cables are of paramount importance.

[0004] The main methods for detecting defects in cable joints currently include:

[0005] Ultrasonic testing: Defects are detected by collecting ultrasonic characteristic parameters of cable joints, such as ultrasonic echo field of view and ultrasonic echo amplitude.

[0006] Partial discharge method: Although it can detect discharge phenomena, it is difficult to pinpoint the discharge source, especially in some complex electrical environments, where the accuracy and reliability of this method may be affected.

[0007] Time Domain Reflectometry (TDR): The TDR method is mainly used to detect physical discontinuities in cables, such as breaks or faulty connectors. However, its detection effect may not be ideal for internal defects caused by insulation aging or moisture.

[0008] The aforementioned existing technologies, whether ultrasonic testing, partial discharge signal detection, or image-based defect detection, are all less efficient than relying on manual image interpretation. Manual inspection takes about 15 minutes per joint, has a high rate of missed detection, with an average missed detection rate of 12% to 18%, and single-view DR imaging is difficult to detect three-dimensional defects, such as bubble volume measurement error of ±23%. Existing automated inspection systems have an accuracy of less than 65% in identifying defects <0.5mm. Summary of the Invention

[0009] In view of the aforementioned existing problems, the present invention is proposed.

[0010] Therefore, this invention provides a cable joint defect detection method based on X-ray imaging and deep learning. By employing dual-energy X-ray data acquisition, multimodal image fusion, and intelligent defect diagnosis, the detection accuracy and coverage of cable joints can be effectively improved, as well as the detection efficiency and stability.

[0011] To address the aforementioned technical problems, this invention provides the following technical solution: a cable joint defect detection method based on X-ray imaging and deep learning, comprising: acquiring multi-energy and multi-angle X-ray images through preset scanning parameters; performing scattering correction and fusion processing on the acquired X-ray images; reconstructing a three-dimensional model of the cable joint based on multi-angle projection data; and extracting defect features using a multi-scale convolutional neural network and outputting the detection results.

[0012] As a preferred embodiment of the cable joint defect detection method based on X-ray imaging and deep learning described in this invention, the acquisition of multi-energy and multi-angle X-ray images includes adjusting the geometric magnification of the X-ray source and the flat panel detector.

[0013] Select X-ray source parameters based on the cable insulation material, including tube voltage, tube current, and exposure time;

[0014] Multi-angle rotation scanning was performed using a 3D PTZ control cable connector, and sequence images were acquired in low, medium, and high energy modes.

[0015] As a preferred embodiment of the cable joint defect detection method based on X-ray imaging and deep learning described in this invention, the scattering correction and fusion processing includes adaptive bilateral filtering for noise reduction of each frame of image.

[0016] The image was decomposed into reflectance and illumination components using an improved Retinex scattering correction method.

[0017] Weighted fusion of multi-energy images from the same angle is performed to generate an enhanced fixed-angle image.

[0018] As a preferred embodiment of the cable joint defect detection method based on X-ray imaging and deep learning described in this invention, the method for reconstructing the three-dimensional model of the cable joint includes filtering and back-projecting multi-angle projection data using a back-projection algorithm.

[0019] The continuous fault images are interpolated into a regular three-dimensional voxel grid to generate three-dimensional volume data for three-dimensional visualization.

[0020] As a preferred embodiment of the cable joint defect detection method based on X-ray imaging and deep learning described in this invention, the defect feature extraction includes extracting three-dimensional geometric features and two-dimensional grayscale features through a multi-scale convolutional neural network.

[0021] By integrating three-dimensional geometric features and two-dimensional grayscale features, defect type, location coordinates, and confidence score are generated.

[0022] The detection results are dynamically matched with the defect feature database, and the matching threshold is updated accordingly.

[0023] As a preferred embodiment of the cable joint defect detection method based on X-ray imaging and deep learning described in this invention, the dynamic matching includes using the DBSCAN clustering algorithm to calculate the distance between new defect features and historical data cluster centers.

[0024] When the distance is less than or equal to a preset threshold, it is classified as a known defect type; when the distance is less than the preset threshold, it is marked as a novel defect and manual review is triggered.

[0025] As a preferred embodiment of the cable joint defect detection method based on X-ray imaging and deep learning described in this invention, the defect feature database includes: uploading detected novel defect samples to a cloud server and marking them as data to be verified; and using an incremental learning algorithm to iteratively optimize the weights of a multi-scale convolutional neural network using the data to be verified.

[0026] Based on the DBSCAN clustering algorithm, the defect features in the cloud database are dynamically clustered, and the cluster centers and matching thresholds of historical data are updated.

[0027] Regularly download updated defect feature data and optimized network weights from the cloud server and synchronize them to the local database.

[0028] Another objective of this invention is to provide a cable joint defect detection system based on X-ray imaging and deep learning.

[0029] As a preferred embodiment of the cable joint defect detection system based on X-ray imaging and deep learning described in this invention, it includes: an acquisition and control module, an image processing module, a 3D modeling module, a defect analysis module, and a database management module;

[0030] The acquisition control module is used to control the three-dimensional pan-tilt unit to perform multi-angle rotation scanning and acquire sequential X-ray images in low, medium and high energy modes;

[0031] The image processing module is used to perform adaptive bilateral filtering denoising, improved Retinex scattering correction, and multi-energy weighted fusion on the acquired images.

[0032] The 3D modeling module is used to generate a 3D voxel mesh through a back projection algorithm and to achieve 3D visualization.

[0033] The defect analysis module is used to extract and fuse three-dimensional geometric features and two-dimensional grayscale features through a multi-scale convolutional neural network, and output the defect type, location coordinates and confidence score.

[0034] The database management module is used to realize the uploading of new defect samples, cloud incremental learning to optimize weights, dynamic clustering to update thresholds, and local data synchronization.

[0035] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a cable joint defect detection method based on X-ray imaging and deep learning.

[0036] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a method for detecting defects in cable joints based on X-ray imaging and deep learning.

[0037] The beneficial effects of this invention are: achieving 0.1mm through multi-angle scanning (36-view) and back-projection algorithms. 3 Voxel resolution 3D reconstruction can accurately calculate crack direction and pore volume (such as detecting pores with a diameter of 0.3 mm), with an error of ≤5%, meeting the power industry's quantitative standards for minute defects; the use of multi-energy exposure (80 / 120 / 200kV combination) and weighted fusion algorithm improves the contrast between insulation delamination and metal burrs by 300%.

[0038] By employing a multi-scale convolutional neural network to fuse 2D grayscale and 3D geometric features, subpixel-level defect localization (error ≤ 0.2mm) is supported. Simultaneously, a three-dimensional spatial consistency constraint loss function is introduced, increasing the insulation layer identification accuracy from 72% to 97%. The detection speed is fast; the digital imaging system allows for rapid acquisition of detection results, contributing to improved detection efficiency, reduced power outage time, and less impact on power grid operation. Attached Figure Description

[0039] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of a cable joint defect detection method based on X-ray imaging and deep learning, provided as an embodiment of the present invention. Detailed Implementation

[0041] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0042] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a method for detecting defects in cable joints based on X-ray imaging and deep learning, including:

[0043] S1: Acquire multi-energy and multi-angle X-ray images using preset scanning parameters.

[0044] Furthermore, an X-ray device is set up and its transmission power is adjusted appropriately; a background data receiving terminal is set up, which is an intelligent device such as a computer with data processing capabilities, and an adaptive scanning parameter optimization module is established within the terminal.

[0045] A mathematical model of tube voltage versus material thickness is established, which can be expressed by the following formula:

[0046] V opt =20·e 0.015δ +65

[0047] Where δ represents the insulation layer thickness in mm, and the experimental verification error is <±3kV. optThe optimal tube voltage is given by e, which is the base of the natural logarithm and approximately equal to 2.71828.

[0048] Dynamic current regulation algorithm:

[0049]

[0050] Where I adj The adjusted dynamic current (unit: mA), V opt For the optimal tube voltage (unit: kV), R eq Where is the equivalent resistance (unit: Ω), and D is the defect feature size.

[0051] The adaptive scanning parameter module inputs the image data obtained by the X-ray equipment into the terminal for data processing. The material thickness of the scanned image is obtained by analysis, and at the same time, the dynamic current adjustment algorithm reduces the noise and oscillation generated by the X-ray equipment scanning and cable joints, thereby improving the dynamic performance and stability of data acquisition.

[0052] It should be noted that the high-voltage cable intermediate joint should be fixed to the 3D pan-tilt unit, and the distance between the X-ray source and the flat panel detector should be adjusted to the optimal geometric magnification; the X-ray source parameters should be set and selected according to the cable insulation material (e.g., 80-150kV for cross-linked polyethylene, 50-120kV for ethylene propylene rubber); the tube current should be set to 3-8mA (to avoid overexposure leading to detector saturation); and the exposure time should be adjusted to 100-500ms (to balance with motion blur).

[0053] Select the appropriate energy mode: low energy (80kV) highlights the insulation layer, medium energy (120kV) provides balanced imaging, and high energy (200kV) penetrates the metal shielding layer; set the pan-tilt unit to rotate 360°, with a step angle of 10° (36 sets of viewing angles in total), and an eccentricity angle of ±30° (enhancing 3D feature capture).

[0054] S2: Perform scattering correction and fusion processing on the acquired X-ray images.

[0055] Furthermore, the image data acquired by the X-ray equipment is enhanced.

[0056] Improved Retinex scattering correction is performed, where the image received by Retinex can be decomposed into illumination and reflectivity components, and the image at each pixel location can be represented as:

[0057] I(i,j)=L(i,j)·R(i,j)

[0058] Where R(i,j) represents the reflectivity component, I(i,j) is the received image, and L(i,j) is the illumination component.

[0059] The atmospheric scattering model is defined as follows:

[0060]

[0061] Where I(x) is the input image, J(x) is the fog-free image, t(x) is atmospheric transmission, and A(x) is the intensity of atmospheric light. Treating the low-light image as the illumination component, the formula can be rewritten as:

[0062]

[0063] Image data is enhanced by combining features.

[0064] In one feasible embodiment, image data enhancement can be achieved through Monte Carlo optical transmission based on a physical model. Specifically, a database of photon scattering parameters of the cable material is constructed, including the X-ray absorption coefficient and scattering cross-section of cross-linked polyethylene / ethylene propylene rubber; for each frame of X-ray image, the path distribution of photons penetrating the cable joint is simulated using the Monte Carlo method; a scattering noise probability map is calculated based on the path distribution, and the noise component is subtracted from the original image in reverse; adaptive contrast stretching is performed on the denoised image to compensate for energy attenuation.

[0065] In another feasible embodiment, image data enhancement can also be achieved through deep learning end-to-end scattering correction. Specifically, synthetic training data is generated to simulate X-ray images with different defect types, angles, and energies in a simulation environment, and real scattering noise is added; a U-Net network is trained, taking noisy images as input and outputting scatter-free images (labeled as ideal images in the simulation environment); the acquired X-ray images are input into the trained U-Net, and the corrected images are directly output.

[0066] It should be noted that for sequential image acquisition, the 3D pan-tilt unit is activated and scanned frame by frame at a preset step angle. Three sets of images with different energy (low / medium / high energy) are acquired at each angle, so that the total number of images is 36 angles × 3 energies = 108 frames.

[0067] Real-time quality control is achieved by calculating the signal-to-noise ratio (SNR) of each frame:

[0068]

[0069] The ROI is the cable joint area, which requires an SNR ≥ 35dB.

[0070] S3: Reconstruct a 3D model of a cable joint based on multi-angle projection data.

[0071] Furthermore, adaptive bilateral filtering denoising applies a filter to each frame of the image (parameters: spatial domain kernel σc = 1.5, gray-level domain kernel σs = 15) to eliminate quantum noise and scattering noise;

[0072] Multi-frame image fusion involves weighted fusion of multi-energy images from the same angle.

[0073] I fused =0.3I low +0.5I mid +0.2I high

[0074] Among them, I fused For the fused fixed-angle image, I low For low-energy images, I mid For medium-energy images, I high This is a medium-energy image;

[0075] 3D reconstruction and evaluation are performed using the Filtered Back Projection (FBP) algorithm to reconstruct the 3D model. The algorithm flow is as follows:

[0076] By using multi-angle projection, the X-ray source-detector system is rotated around the object (typically 180°–360°) to acquire two-dimensional projected images from different angles. A high-pass filter is applied to the projected data to compensate for the blurring effect during the back-projection process. This high-pass filter can be expressed by the following mathematical expression:

[0077] P filtered (θ, s) = P(θ, s) * h(s)

[0078] Where P(θ, s) is the original projection at angle θ, and h(s) is the slope filter kernel.

[0079] The filtered projection data is projected back along the original ray path into the image space, and the contributions from all angles are summed.

[0080] Repeat the above steps to reconstruct all faults (Z-axis layers) and stack them into three-dimensional volume data.

[0081] Furthermore, the three-dimensional reconstruction mentioned above also includes:

[0082] Interpolate continuous tomographic images into a regular three-dimensional voxel grid; display internal structures through ray casting; surface rendering: extract isosurfaces for specific thresholds (such as bone density).

[0083] Reconstruction quality is evaluated using the following indicators:

[0084] Spatial resolution: Tests the sharpness of high-contrast edges; high spatial resolution > low spatial resolution.

[0085] Contrast-to-noise ratio: The lower the noise ratio, the better the reconstruction quality;

[0086] Sensitivity in detecting cracks, clarity of tumor boundaries, etc.

[0087] S4: Extract defect features using a multi-scale convolutional neural network and output the detection results.

[0088] Furthermore, by combining multi-scale convolutional neural networks (CNNs) with image processing and deep learning techniques, it is possible to efficiently identify physical defects (such as bubbles and cracks) or thermal anomalies (such as localized overheating) inside the joint. A defect feature database including 12 typical defect types is established, and this database is used to train the system. Simultaneously, a crack detection algorithm based on the Hessian matrix is ​​used for defect identification during detection. The crack detection algorithm is as follows:

[0089]

[0090] In the formula: λ ratio λ1 and λ2 are the crack feature ratios and pixel intensity feature values, respectively.

[0091] When λratio > 8, it is judged as a linear defect (experiments show an accuracy rate of 98.2%).

[0092] It should be noted that the multi-scale convolutional neural network (MS-CNN) is used to process the input image data and the defect feature database to extract and train the defect features. The 3D convolutional layer extracts three-dimensional geometric features, such as crack direction and pore depth, the 2D convolutional layer extracts two-dimensional grayscale features, such as insulation layering and metal burrs, and the fusion layer performs weighted fusion branch feature extraction.

[0093] The obtained defects are classified by defect type (porosity, crack, stress cone misalignment, etc.), location coordinates (positioning error in three-dimensional space ≤ 0.2 mm), and confidence score (threshold ≥ 0.9 is used to determine a valid defect), and the defect results are output.

[0094] Dynamic database matching compares the detection results with a defect feature database and uses the DBSCAN clustering algorithm to update the matching threshold.

[0095] (1) If the distance between the new defect feature and the cluster center of the historical data is ≤0.1, then it is classified as a known type;

[0096] (2) If the distance is >0.1, mark it as a new type of defect and trigger manual review.

[0097] After acquiring defect samples, the software model can be updated in real time. The latest defect feature library is downloaded from the cloud every month, and the MS-CNN weights are incrementally learned and updated to improve the accuracy of defect identification of cable joints. Risk assessment and reporting can also be performed sequentially.

[0098] In a feasible embodiment, crack detection can be achieved through morphological skeleton analysis and random forest classification. Specifically, multi-scale morphological operations are performed on the 3D reconstructed tomographic image to separate the background from potential crack regions; the topological skeleton of the crack candidate region is extracted, and geometric features such as skeleton branch length, curvature, and connectivity are calculated; the geometric features are input into a pre-trained random forest classifier, which outputs the crack probability (threshold > 0.95 is judged as a crack); and the crack regions in the continuous fault are labeled with 3D connected components to remove isolated noise points.

[0099] In another feasible embodiment, crack detection can also be achieved through a three-dimensional convolutional attention mechanism. Specifically, a 3D crack annotation dataset is constructed, and crack voxels in the three-dimensional volume data are manually annotated; 3DAttentionU-Net is trained, with the network input being the three-dimensional volume data and the output being a crack voxel probability map; a channel attention module is added to the encoder to enhance the feature response to linear structures; and during deployment, the binary crack segmentation result is directly output (probability > 0.9 is judged as crack voxel).

[0100] Example 2 is the second embodiment of the present invention, which differs from the previous embodiment in that:

[0101] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0102] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0103] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0104] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0105] Example 3, an embodiment of the present invention, provides a cable joint defect detection system based on X-ray imaging and deep learning, including an acquisition and control module, an image processing module, a 3D modeling module, a defect analysis module, and a database management module;

[0106] The acquisition and control module is used to control the 3D pan-tilt unit to perform multi-angle rotation scanning and acquire sequential X-ray images in low, medium and high energy modes;

[0107] The image processing module is used to perform adaptive bilateral filtering for noise reduction, improved Retinex scattering correction, and multi-energy weighted fusion on the acquired images.

[0108] The 3D modeling module is used to generate 3D voxel meshes using a back-projection algorithm and to achieve 3D visualization.

[0109] The defect analysis module is used to extract and fuse three-dimensional geometric features and two-dimensional grayscale features through a multi-scale convolutional neural network, and output the defect type, location coordinates and confidence score.

[0110] The database management module is used to upload new defect samples, optimize weights through cloud-based incremental learning, update thresholds through dynamic clustering, and synchronize local data.

[0111] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for detecting defects of cable intermediate joint based on X-ray imaging and deep learning, characterized in that: The method comprises the steps of: acquiring X-ray images of multiple energies and multiple angles by presetting scanning parameters; performing scatter correction and fusion processing on the acquired X-ray images; reconstructing a three-dimensional model of the cable intermediate joint based on the multiple-angle projection data; extracting defect features by using a multi-scale convolutional neural network and outputting detection results.

2. The X-ray imaging and deep learning based cable joint defect detection method of claim 1, wherein: The acquisition of X-ray images of multiple energies and multiple angles comprises adjusting the geometric magnification of the X-ray source and the flat panel detector; selecting X-ray source parameters, including tube voltage, tube current and exposure time, according to the cable insulation material; performing multi-angle rotational scanning on the cable intermediate joint by a three-dimensional gimbal, and acquiring sequence images under low, medium and high energy modes.

3. The X-ray imaging and deep learning based cable joint defect detection method of claim 2, wherein: The scatter correction and fusion processing comprises self-adaptive bilateral filter denoising for each frame of image; decomposing the image by an improved Retinex scatter correction into reflectance component and illumination component; performing weighted fusion on multiple-energy images of the same angle to generate enhanced fixed-angle images.

4. The X-ray imaging and deep learning based cable joint defect detection method of claim 3, wherein: The reconstruction of the three-dimensional model of the cable intermediate joint comprises filtering and inverse projection of the multiple-angle projection data by a back-projection algorithm; interpolating the continuous tomographic images into a regular three-dimensional voxel grid to generate three-dimensional volume data and perform three-dimensional visualization.

5. The X-ray imaging and deep learning based cable joint defect detection method of claim 4, wherein: The extraction of defect features comprises extracting three-dimensional geometric features and two-dimensional gray-scale features by a multi-scale convolutional neural network; fusing the three-dimensional geometric features and the two-dimensional gray-scale features to generate defect types, position coordinates and confidence scores; performing dynamic matching of the detection results with the defect feature database and updating the matching threshold.

6. The X-ray imaging and deep learning based cable intermediate joint defect detection method of claim 5, wherein: The dynamic matching comprises calculating the distance between the new defect features and the historical data clustering center by using a DBSCAN clustering algorithm; when the distance is less than or equal to the preset threshold, the new defect is classified as a known defect type; when the distance is less than the preset threshold, the new defect is marked as a new type of defect and triggers manual review.

7. The X-ray imaging and deep learning based cable joint defect detection method of claim 6, wherein: The defect feature database comprises uploading the detected new defect samples to a cloud server and marking them as to-be-verified data, and iteratively optimizing the weights of the multi-scale convolutional neural network by using the to-be-verified data through an incremental learning algorithm; performing dynamic clustering of the defect features in the cloud database based on the DBSCAN clustering algorithm, and updating the historical data clustering center and the matching threshold; periodically downloading the updated defect feature data and the optimized network weights from the cloud server and synchronizing them to the local database.

8. A system employing the method for detecting defects in cable intermediate joint based on X-ray imaging and deep learning according to any one of claims 1 to 7, characterized in that: The system comprises a collection control module, an image processing module, a three-dimensional modeling module, a defect analysis module and a database management module; The collection control module is configured to control the three-dimensional gimbal to perform multi-angle rotational scanning and acquire sequence X-ray images under low, medium and high energy modes. The image processing module is configured to perform self-adaptive bilateral filter denoising, improved Retinex scatter correction and multi-energy weighted fusion on the acquired images. The three-dimensional modeling module is configured to generate a three-dimensional voxel grid by a back-projection algorithm and realize three-dimensional visualization. The defect analysis module is configured to extract and fuse three-dimensional geometric features and two-dimensional gray-scale features by a multi-scale convolutional neural network, and output defect types, position coordinates and confidence scores. The database management module is used for uploading new defect samples, optimizing weights by cloud incremental learning, dynamically updating thresholds by clustering, and synchronizing local data. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to realize the steps of the cable intermediate joint defect detection method based on X-ray imaging and deep learning in any one of claims 1-7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to realize the steps of the cable intermediate joint defect detection method based on X-ray imaging and deep learning in any one of claims 1-7.

Images