Defect detection method and device for outer semi-conductive layer of high-voltage cable

Abstract

The invention discloses a method and a device for detecting defects of an outer semi-conductive layer of a high-voltage cable. The method and the device are used for solving the technical problem that existing cutting equipment for the outer semi-conductive layer of the high-voltage cable lacks defect detection on the high-voltage cable. The method comprises the following steps: acquiring a to-be-detected image of an outer semi-conductive layer of a high-voltage cable; performing image reconstruction on the to-be-detected image to obtain a reconstructed detection image; performing feature recognition and extraction combining depth separable convolution and multi-scale feature fusion on the reconstructed detection image to obtain key image features; and performing anomaly detection based on defect types and positions on the key image features, and outputting a defect detection result of the outer semi-conductive layer of the high-voltage cable. Therefore, based on machine vision, in combination with depth separable convolution and multi-scale feature fusion, recognition and extraction of key image features are realized, anomaly detection is performed based on defect types and positions, and defect detection of the outer semi-conductive layer of the high-voltage cable can be quickly and accurately realized.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment detection, and in particular to a method for detecting defects in an outer semi-conductive layer of a high-voltage cable, a device for detecting defects in an outer semi-conductive layer of a high-voltage cable, an electronic device and a storage medium. Background Art

[0002] High voltage cable refers to a power cable used to transmit power between 1kv and 1000kv (unit: kilovolt). It is mostly used in power transmission and distribution. With the widespread application of high voltage cables, the research and development of high voltage cable outer semi-conductive layer cutting equipment has achieved remarkable results. Although the existing high voltage cable outer semi-conductive layer cutting equipment can cut the outer semi-conductive layer of high voltage cables, it still has many shortcomings.

[0003] First, the existing cable cutting equipment is mainly for single cable entry. The outer semi-conductive layer of the high-voltage cable is stripped under the action of the cutting tool, and the stripping efficiency is not high. And it is easy to have deviations, resulting in scratches, pits and other defects on the outer semi-conductive layer of the high-voltage cable.

[0004] Secondly, the existing cable cutting equipment has a relatively simple cutting equipment structure, so it is difficult to make the outer semi-conductive layer of the high-voltage cable meet the processing accuracy requirements through a single cutting operation. The high-voltage cable that has been cut by the high-voltage cable outer semi-conductive layer cutting equipment will have defects such as cracks and scratches caused by the cutting effect, and the processing accuracy fails to meet the requirements. This will not only affect the fit between the cable conductor and the crimping tube of the high-voltage cable. It is also very likely to reduce the service life and transmission efficiency of the high-voltage cable. It may even cause serious safety hazards and have a serious impact on public safety and economic benefits.

[0005] The existing high-voltage cable outer semi-conductive layer cutting equipment lacks the defect detection of high-voltage cables. Therefore, it is of great practical significance to study a method for detecting defects in the outer semi-conductive layer of high-voltage cables. Summary of the invention

[0006] The present invention provides a method for detecting defects in the outer semi-conductive layer of a high-voltage cable, a device for detecting defects in the outer semi-conductive layer of a high-voltage cable, an electronic device and a storage medium, which are used to solve or partially solve the technical problem that existing high-voltage cable outer semi-conductive layer cutting equipment lacks defect detection for high-voltage cables.

[0007] The present invention provides a method for detecting defects in an outer semi-conductive layer of a high-voltage cable, the method comprising:

[0008] Acquire the image to be inspected of the outer semi-conductive layer of the high-voltage cable;

[0009] Reconstructing the image to be detected to obtain a reconstructed detection image;

[0010] Performing feature recognition and extraction combining depth-separable convolution with multi-scale feature fusion on the reconstructed detection image to obtain key image features;

[0011] The key image features are subjected to abnormality detection based on defect categories and positions, and defect detection results of the outer semi-conductive layer of the high-voltage cable are output.

[0012] Optionally, reconstructing the image to be detected to obtain a reconstructed detection image includes:

[0013] Based on image graying, converting the image to be detected into a grayscale image;

[0014] Removing image noise from the grayscale image by using a Gaussian blur filter to obtain a denoised image;

[0015] The denoised image is converted into a binary image by image binarization, and a target area of ​​the binary image is separated by threshold segmentation to obtain a reconstructed detection image.

[0016] Optionally, the step of performing feature recognition and extraction on the reconstructed detection image by combining depth-separable convolution with multi-scale feature fusion to obtain key image features includes:

[0017] Inputting the reconstructed detection image into a pre-trained anomaly detection model;

[0018] According to the anomaly detection model, combined with an edge detection algorithm, identifying and extracting a preliminary detection feature map of the reconstructed detection image;

[0019] Performing depth-separable convolution on the preliminary detection feature map to obtain a depth-convolution feature map;

[0020] Inputting the depth convolution feature map into a plurality of first convolution layers of different scales for convolution processing respectively, obtaining convolution scale feature maps corresponding to each of the first convolution layers, and fusing the convolution scale feature maps with each other in a top-down and bottom-up manner to achieve multi-scale feature fusion, and obtaining a multi-scale fused feature map;

[0021] Perform residual convolution processing on the multi-scale fusion feature map to obtain key image features.

[0022] Optionally, performing abnormality detection on the key image features based on defect category and position, and outputting defect detection results of the outer semi-conductive layer of the high-voltage cable, comprises:

[0023] Performing anomaly detection based on defect category and location on the key image features;

[0024] When at least one abnormal feature is detected, the defect category closest to each abnormal feature is determined respectively, and a defect optimization detection frame of a corresponding size is generated at the location of the defect category. The defect optimization detection frame is obtained by optimizing defect category recognition and position positioning based on a multi-task loss function during the training of the abnormality detection model;

[0025] Comparing the abnormal feature information corresponding to each defect detection frame with the preset defect feature information respectively, so as to determine whether an abnormal defect occurs in the outer semi-conductive layer of the high-voltage cable;

[0026] Based on the feature comparison result, the defect detection result of the outer semi-conductive layer of the high-voltage cable is output.

[0027] Optionally, the training process of the anomaly detection model includes:

[0028] Collect multiple defect images of different defects in the outer semi-conductive layer of high-voltage cables;

[0029] Performing data preprocessing on the plurality of defect images and performing image reconstruction to obtain a reconstructed defect image data set;

[0030] Dividing the reconstructed defect image data set into an image defect training set and an image defect verification set according to a preset ratio;

[0031] An initial anomaly detection model is constructed using an adaptive convolutional network including an auxiliary classifier as the network architecture; the initial anomaly detection model includes a feature extraction module, a multi-scale feature fusion module and a defect detection module; wherein:

[0032] The multi-scale feature fusion module adopts a feature pyramid network, and the feature pyramid network includes multiple first convolutional layers, and the multiple first convolutional layers are used to generate feature maps of different scales;

[0033] The feature extraction module includes at least one second convolutional layer and multiple third convolutional layers, the at least one second convolutional layer is used to perform preliminary feature extraction on the image of the input model, and each of the third convolutional layers adopts a convolution type of depthwise separable convolution;

[0034] The defect detection module is composed of a defect detection submodule and a feature comparison submodule. The defect detection submodule is used to perform anomaly detection on the key image features output by the multi-scale feature fusion module, and generate at least one defect detection frame based on the feature anomaly detection result, and the defect detection frame corresponds to the candidate detection area; the feature comparison submodule is used to compare and verify the abnormal feature information in the candidate detection area, and generate a defect detection result;

[0035] Performing parameter adjustment on each of the first convolutional layer, the second convolutional layer, and the third convolutional layer, wherein the parameter adjustment includes setting a convolution kernel size, a step size, and padding;

[0036] Based on the defined multi-task loss function, the initial anomaly detection model is trained by the image defect training set, and the trained anomaly detection model is verified by the image defect verification set;

[0037] In the process of model training and verification, based on the model verification and evaluation results, the gradient descent method is used to iteratively optimize and update the parameters of the multi-task loss function to optimize the defect detection box's ability to identify defect categories and locate defect locations;

[0038] When the number of iterations reaches the preset maximum number of iterations, or the model prediction error is less than or equal to the preset error threshold, the iteration is stopped and the final anomaly detection model is output.

[0039] Optionally, the multi-task loss function is formed by comprehensively considering the position error, position weighting, classification error and category weighting of the defect detection frame; the position weighting is used to evaluate the shape and position change of the defect detection frame to reflect the importance of position positioning; the category weighting is used to evaluate the defect category of the abnormal features in the defect detection frame to reflect the importance of category identification; the specific expression of the multi-task loss function is as follows:

[0040]

[0041] in, represents the multi-task loss function, representing the total loss; and Represent the total weight of category weighting and position weighting respectively; and denote the category-weighted and position-weighted losses respectively.

[0042] Optionally, after outputting the defect detection result of the outer semiconductive layer of the high-voltage cable, the method further comprises:

[0043] When the defect detection result indicates that an abnormal defect occurs in the outer semi-conductive layer of the high-voltage cable, a real-time alarm is triggered for the abnormal defect based on a real-time alarm mechanism, and the defect detection result is sent to the management end of the cutting equipment corresponding to the outer semi-conductive layer of the high-voltage cable.

[0044] The present invention also provides a device for detecting defects in an outer semi-conductive layer of a high-voltage cable, comprising:

[0045] A data acquisition unit, used for acquiring an image to be detected of the outer semi-conductive layer of the high-voltage cable;

[0046] An image reconstruction unit, used for reconstructing the image to be detected to obtain a reconstructed detection image;

[0047] A feature extraction unit, used to perform feature recognition and extraction on the reconstructed detection image by combining deep separable convolution with multi-scale feature fusion to obtain key image features;

[0048] The anomaly detection unit is used to perform anomaly detection on the key image features based on defect type and position, and output defect detection results of the outer semi-conductive layer of the high-voltage cable.

[0049] The present invention also provides an electronic device, the device comprising a processor and a memory:

[0050] The memory is used to store program code and transmit the program code to the processor;

[0051] The processor is used to execute the method for detecting defects in the outer semiconductive layer of a high-voltage cable as described in any one of the above items according to the instructions in the program code.

[0052] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store program codes, and the program codes are used to execute the method for detecting defects in the outer semiconductive layer of a high-voltage cable as described in any one of the above items.

[0053] It can be seen from the above technical solutions that the present invention has the following advantages:

[0054] A method for detecting defects in the outer semi-conductive layer of a high-voltage cable is provided. First, an image to be detected of the outer semi-conductive layer of the high-voltage cable is obtained; then, the image to be detected is reconstructed to obtain a reconstructed detection image; then, feature recognition and extraction combining deep separable convolution with multi-scale feature fusion are performed on the reconstructed detection image to obtain key image features; finally, anomaly detection based on defect category and position is performed on the key image features, and the defect detection result of the outer semi-conductive layer of the high-voltage cable is output. Thus, based on machine vision, deep separable convolution and multi-scale feature fusion are combined to realize recognition and extraction of key image features, and anomaly detection is performed based on defect category and position, so that defect detection of the outer semi-conductive layer of the high-voltage cable can be realized quickly and accurately. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 A flowchart of a method for detecting defects in the outer semi-conductive layer of a high-voltage cable;

[0057] Figure 2 A flowchart of the training process of an anomaly detection model;

[0058] Figure 3 A schematic diagram of a process for optimizing model parameters by introducing an optimized loss function;

[0059] Figure 4 It is a schematic diagram of the overall process of a method for detecting defects in the outer semi-conductive layer of a high-voltage cable;

[0060] Figure 5 The present invention is a structural block diagram of a device for detecting defects in the outer semi-conductive layer of a high-voltage cable. DETAILED DESCRIPTION

[0061] The embodiments of the present invention provide a method for detecting defects in the outer semi-conductive layer of a high-voltage cable, a device for detecting defects in the outer semi-conductive layer of a high-voltage cable, an electronic device and a storage medium, which are used to solve or partially solve the technical problem that existing high-voltage cable outer semi-conductive layer cutting equipment lacks defect detection for high-voltage cables.

[0062] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] As an example, although the existing high-voltage cable outer semi-conductive layer cutting equipment can cut the outer semi-conductive layer of the high-voltage cable, it still has many shortcomings.

[0064] First, the existing cable cutting equipment is mainly for single cable entry. The outer semi-conductive layer of the high-voltage cable is stripped under the action of the cutting tool, and the stripping efficiency is not high. And it is easy to have deviations, resulting in scratches, pits and other defects on the outer semi-conductive layer of the high-voltage cable.

[0065] Secondly, the existing cable cutting equipment has a relatively simple cutting equipment structure, so it is difficult to make the outer semi-conductive layer of the high-voltage cable meet the processing accuracy requirements through a single cutting operation. The high-voltage cable that has been cut by the high-voltage cable outer semi-conductive layer cutting equipment will have defects such as cracks and scratches caused by the cutting effect, and the processing accuracy fails to meet the requirements. This will not only affect the fit between the cable conductor and the crimping tube of the high-voltage cable. It is also very likely to reduce the service life and transmission efficiency of the high-voltage cable. It may even cause serious safety hazards and have a serious impact on public safety and economic benefits.

[0066] The existing high-voltage cable outer semi-conductive layer cutting equipment lacks the defect detection of high-voltage cables. Therefore, it is of great practical significance to study a method for detecting defects in the outer semi-conductive layer of high-voltage cables.

[0067] Therefore, one of the core invention points of the embodiment of the present invention is to provide a method and device for detecting defects in the outer semi-conductive layer of a high-voltage cable based on machine vision. (1) First, by collecting images of the outer semi-conductive layer of the high-voltage cable to be inspected, and based on image preprocessing and feature extraction, the clarity of the original image is significantly improved, the features of interest are enhanced, and electronic noise and optical interference are reduced, which is conducive to the detection and identification of the outer semi-conductive layer of the high-voltage cable. (2) Secondly, an improved adaptive convolutional network is adopted, and by combining deep separable convolution, multi-scale feature fusion processing and hierarchical feature learning, the key image features of the outer semi-conductive layer of the high-voltage cable are automatically learned, the detection accuracy of the model for defects is improved, and errors caused by human intervention and subjective judgment are avoided. (3) At the same time, the optimization loss function is introduced to perform abnormal detection on the outer semi-conductive layer of the high-voltage cable to improve the performance of the model, thereby more accurately identifying the defects of the outer semi-conductive layer of the high-voltage cable, adapting to image detection of different sizes and angles, and avoiding the situation of missing or misdetecting small target defects, further improving the accuracy and stability of detection. (4) Through the real-time alarm mechanism, a real-time alarm is triggered based on the abnormal defect detection results, and the relevant detection results are sent to the management end of the high-voltage cable outer semi-conductive layer cutting equipment to achieve real-time monitoring of the effect of the high-voltage cable outer semi-conductive layer cutting equipment, so as to timely discover and effectively deal with abnormal conditions and improve the reliability and stability of the cutting equipment.

[0068] The defect detection method and device proposed in the present invention can be integrated into the cutting equipment of the outer semi-conductive layer of a high-voltage cable to assist and detect the operating effect of the cutting equipment. In general, the technical solution of the present invention has the advantages of high automation, high detection accuracy, and good real-time performance, and can effectively improve the efficiency and accuracy of defect detection of the outer semi-conductive layer of a high-voltage cable. By adopting this solution, the defect detection of the outer semi-conductive layer of a high-voltage cable can be made more intelligent, efficient, and precise. This provides a reliable quality control method for the cutting equipment of the outer semi-conductive layer of a high-voltage cable, and improves the overall operating effect of the cutting equipment.

[0069] Reference Figure 1 , shows a flowchart of a method for detecting defects in the outer semi-conductive layer of a high-voltage cable provided by an embodiment of the present invention, which may specifically include the following steps:

[0070] Step 101, obtaining an image to be inspected of an outer semi-conductive layer of a high-voltage cable;

[0071] In a specific implementation, an image acquisition device (such as a camera) can be used to capture the outer semi-conductive layer of the high-voltage cable to be inspected in real time to obtain a corresponding captured image as the image to be inspected.

[0072] Step 102, reconstructing the image to be detected to obtain a reconstructed detection image;

[0073] Specifically, the process of reconstructing the image to be detected to obtain the reconstructed detection image can be achieved by executing the following sub-steps S01 to S03:

[0074] Step S01: based on image grayscale conversion, convert the image to be detected into a grayscale image;

[0075] The purpose of image grayscale is to convert the image into a grayscale image to reduce the amount of processed data and focus on the texture and structure of the image.

[0076] Step S02: removing the image noise of the grayscale image by using a Gaussian blur filter to obtain a denoised image;

[0077] Gaussian blur filter can be used to remove image noise to achieve the purpose of image denoising.

[0078] Step S03: converting the denoised image into a binary image by image binarization, and separating the target area of ​​the binary image by threshold segmentation to obtain a reconstructed detection image.

[0079] Image binarization mainly converts the image into a binary image by setting a threshold so as to more clearly distinguish the target from the background.

[0080] Image segmentation uses threshold segmentation image segmentation technology to separate the target area.

[0081] Step 103, performing feature recognition and extraction combining depthwise separable convolution and multi-scale feature fusion on the reconstructed detection image to obtain key image features;

[0082] In practical applications, an intelligent model for detecting anomalies in images of the outer semi-conductive layer of a high-voltage cable, namely, an anomaly detection model, can be first trained based on machine vision.

[0083] Figure 2 A schematic diagram of a training process flow of an anomaly detection model provided by an embodiment of the present invention is shown.

[0084] Combination Figure 2 In a specific implementation, the anomaly detection model can be trained by executing the following steps S11 to S18:

[0085] Step S11: collecting multiple defect images of different defects of the outer semi-conductive layer of the high-voltage cable;

[0086] First, a large amount of image data of different defects of the outer semi-conductive layer of the high-voltage cable needs to be collected. The outer semi-conductive layer of the high-voltage cable is photographed by an image acquisition device (such as a camera) to obtain image acquisition results of different defects of the outer semi-conductive layer of the high-voltage cable.

[0087] Among them, the outer semi-conductive layer of the high-voltage cable is the outer semi-conductive layer of the high-voltage cable after being operated by the outer semi-conductive layer cutting equipment of the high-voltage cable.

[0088] Then the image acquisition results are annotated. The defect locations and defect types (cable defect categories caused by different damages) of the outer semi-conductive layer of the high-voltage cable are mainly annotated as defect image datasets for model training. The defect image dataset contains multiple annotated defect images.

[0089] Step S12: performing data preprocessing on the multiple defect images and performing image reconstruction to obtain a reconstructed defect image data set;

[0090] In order to improve the prediction accuracy and feasibility of the model and reduce the amount of calculation, the collected data set needs to be preprocessed and image reconstructed before formal model training.

[0091] Specifically, OpenCV (Open Source Computer Vision Library) can be used to preprocess the defect image, and reconstruct the image to reduce the electronic noise and optical interference mixed in the image acquisition, improve the clarity of the original image, and enhance the features of interest in the image to facilitate the detection and identification of defects. Preprocessing operations include deleting repeated redundant information, filling in vacant values, deleting erroneous information, etc. Reconstructing the image can refer to the above-mentioned related steps and will not be repeated here.

[0092] Step S13: dividing the reconstructed defect image data set into an image defect training set and an image defect verification set according to a preset ratio;

[0093] The reconstructed defect image data set can be divided into an image defect training set and an image defect verification set according to a preset ratio of 7:3 or 8:2 (or other suitable division ratios) for subsequent model training and verification optimization.

[0094] Step S14: constructing an initial anomaly detection model using an adaptive convolutional network including an auxiliary classifier as a network architecture;

[0095] In the embodiment of the present invention, the adopted anomaly detection network structure is improved by introducing new convolution blocks, improving convolution layers, multi-scale feature fusion and using residual connections. The main improvements involved are as follows:

[0096] Introducing a new convolutional block: Introducing an adaptive convolutional neural network structure ACNet (Adaptive Convolutional Network, a network architecture using an adaptive convolution mechanism) that includes an auxiliary classifier. The performance of the main detector is enhanced by introducing an auxiliary classifier to improve detection accuracy, enhance robustness, and improve recall. ACNet is used as the basic network architecture to improve the flexibility and expressiveness of convolutional neural networks through adaptive convolution, thereby achieving better performance in image classification and object detection.

[0097] Improved convolutional layers: Set multiple different convolutional layers in the feature extraction and feature fusion processes, and adjust parameters such as convolution kernel size, step size, and padding. At the same time, some convolutional layers in the feature extraction module use the convolution type of depthwise separable convolution to increase feature expression capabilities, improve model robustness, and improve detection accuracy.

[0098] Multi-scale feature fusion: The feature pyramid network is used to achieve multi-scale feature fusion. The feature pyramid network contains multiple convolutional layers of different scales. Each convolutional layer generates feature maps of different scales. These feature maps are fused with each other in a top-down and bottom-up manner to achieve multi-scale information fusion. Through multi-scale feature fusion, positioning accuracy can be improved, detail information can be enhanced, and segmentation effects can be improved.

[0099] Use residual connections: Refer to ResNet (Residual Network, a convolutional neural network architecture / residual network in deep learning), and introduce residual blocks after multi-scale feature fusion. The residual block consists of two or three convolutional layers, and a skip connection is added between the input and output. The skip connection allows information and gradients to be passed directly from the previous layer to the next layer. When performing feature processing, the residual between the input and output of the residual block can be added to the output, and the skip connection can be used to alleviate the gradient vanishing and gradient exploding, accelerate training, and improve model performance.

[0100] Improved region proposal network: The candidate region generation and detection process are integrated into one network as the region proposal network (i.e., defect detection module). By integrating the modules, the detection process can be simplified and end-to-end training can be allowed to detect objects directly from the original image.

[0101] Improved target detection algorithm: Combined with multi-scale feature fusion and optimized loss function, multiple scale network training is performed for target detection, so that the algorithm model can better detect targets of different sizes. For data set samples, the data set can also be diversified by rotating, scaling, cropping, changing the color of the data image, etc., to improve the generalization ability of the algorithm model.

[0102] Combined with the above content, the initial anomaly detection model can mainly include a feature extraction module, a multi-scale feature fusion module and a defect detection module.

[0103] The multi-scale feature fusion module uses a feature pyramid network. The feature pyramid network contains multiple first convolutional layers. Multiple first convolutional layers are used to generate feature maps of different scales.

[0104] The feature extraction module includes at least one second convolution layer and multiple third convolution layers. At least one second convolution layer is used to perform preliminary feature extraction on the image of the input model. Each third convolution layer adopts a convolution type of depth-separable convolution.

[0105] A residual block can also be set between the feature extraction module and the defect detection module to perform residual convolution processing on the features output by the feature extraction module to alleviate the gradient disappearance and gradient explosion problems.

[0106] The defect detection module consists of a defect detection submodule and a feature comparison submodule. The defect detection submodule is used to perform anomaly detection on the key image features output by the multi-scale feature fusion module, and generate at least one defect detection frame based on the feature anomaly detection results. The defect detection frame corresponds to the candidate detection area. The feature comparison submodule is used to compare and verify the abnormal feature information in the candidate detection area, perform feature comparison on the abnormal feature information in the candidate detection area and the expected input defect feature information, and generate defect detection results.

[0107] Step S15: adjusting parameters of the initial detection model;

[0108] Specifically, parameters of each first convolution layer, second convolution layer, and third convolution layer are adjusted, and the parameter adjustment includes setting the convolution kernel size, step size, and padding;

[0109] Step S16: Based on the defined multi-task loss function, the initial anomaly detection model is trained using the image defect training set, and the trained anomaly detection model is verified using the image defect verification set;

[0110] During the model training and verification process, the improved adaptive convolutional neural network introduced earlier is used to identify and extract features such as edges, corners, and textures in the image of the outer semi-conductive layer of the high-voltage cable, so as to extract the key features of the image of the outer semi-conductive layer of the high-voltage cable.

[0111] Specifically, first, an edge detection algorithm (such as the Canny algorithm) can be used to extract the edge contour of the outer semi-conductive layer of the high-voltage cable; then, based on the edge contour, the shape characteristics of the outer semi-conductive layer of the high-voltage cable, such as area, circularity, etc., are calculated; then, the size characteristics of the outer semi-conductive layer of the high-voltage cable, such as length, width, height, etc., are calculated; the extracted key features such as edge contour, shape and size are used as input for defect detection to carry out subsequent defect detection and classification processes.

[0112] For defect detection, abnormality detection is performed by comparing the key features of the outer semi-conductive layer of the high-voltage cable to determine whether there are abnormal defects in the outer semi-conductive layer of the high-voltage cable in the image.

[0113] In order to make defect detection more targeted and accurate, the embodiment of the present invention introduces an optimized loss function to solve the balance problem of large and small frames when penalizing. Based on the optimized loss function, the model performs better when processing targets of different sizes, more accurately identifies defects of different sizes in the outer semi-conductive layer of the high-voltage cable, and improves the accuracy of the abnormality detection process.

[0114] Among them, the optimized loss function is specifically used to balance the weights between positive and negative samples of different defects during the training process, improve the recognition accuracy of the loss edge, and optimize the performance and stability of the detection model.

[0115] Firstly, the position error, classification error and geometric factors (including position weighting and category weighting) of the defect detection box are comprehensively considered to define the multi-task loss function.

[0116] Among them, the geometric factor is mainly used to evaluate the shape and position changes of the defect detection frame, and to penalize parameters such as the aspect ratio and center point offset of the frame to improve the matching degree between the geometric characteristics of the defect detection frame and the actual defect.

[0117] In multi-task learning, especially for tasks such as object detection that involve category recognition and location estimation, the loss function usually consists of multiple parts. Category weighting and location weighting mean that different parts of the loss function have different weights, reflecting the different importance of category recognition and location positioning.

[0118] During the training process, the gradient descent method is used to optimize the multi-task loss function and dynamically adjust the neural network parameters to adapt to different detection tasks.

[0119] Combined with the previous discussion, it can be concluded that the multi-task loss function is composed of the position error, position weighting, classification error and category weighting of the defect detection frame. Among them, the position weighting is used to evaluate the shape and position changes of the defect detection frame to reflect the importance of position positioning. Category weighting is used to evaluate the defect category of abnormal features in the defect detection frame to reflect the importance of category recognition.

[0120] Specifically, the specific expression of the multi-task loss function is as follows:

[0121]

[0122] in, represents the multi-task loss function, representing the total loss; and Represent the total weight of category weighting and position weighting respectively; and denote the category-weighted and position-weighted losses respectively.

[0123] Step S17: In the process of model training and verification, based on the model verification and evaluation results, the gradient descent method is used to iteratively optimize and update the parameters of the multi-task loss function to optimize the defect detection box's ability to identify defect categories and locate defect locations;

[0124] Step S18: When the number of iterations reaches the preset maximum number of iterations, or the model prediction error is less than or equal to the preset error threshold, the iteration is stopped and the final anomaly detection model is output.

[0125] Figure 3 This is a flow chart of introducing an optimization loss function to optimize model parameters.

[0126] Combination Figure 3 , the process of introducing the optimization loss function to optimize the model parameters is as follows:

[0127] The goal of defining the loss function is to reduce the error of the defect detection box, including position error, size error, and classification error. For the detection of defects of different sizes in the outer semi-conductive layer of the high-voltage cable, the loss function is optimized to achieve a balance between defects of different sizes.

[0128] Design loss function: When designing the loss function, it is necessary to consider the above mentioned errors to adapt to defects of different sizes. Here, the cross entropy loss function is used, which can be modified as a regression task for the coordinates and size of the defect detection box (predicted bounding box).

[0129] Combining IOU Loss (Intersection over Union Loss, a commonly used metric and loss function) and Focal Loss (a loss function designed specifically to solve the problem of class imbalance): IOU Loss encourages the bounding box predicted by the model to have a higher overlap with the true bounding box. Focal Loss reduces the weight of simple samples, making the model pay more attention to difficult samples. By combining IOU Loss and Focal Loss, bounding boxes of different sizes are balanced, and different weights are assigned to different types of losses to reflect the importance of different defects.

[0130] Train the network model: The designed loss function is used for model training. During the training process, the model will learn how to predict the location and size of defects in the outer semi-conductive layer of the high-voltage cable based on the input image.

[0131] Parameter Tuning: Evaluate the performance of the model on the validation set and tune the parameters or combination of loss functions as needed.

[0132] Iterative optimization: Assuming that the expected goal of the model is not achieved by adjusting the parameters (the model prediction error is greater than the preset error threshold), the loss function is iteratively optimized. Specifically, through forward propagation, the model output and predicted value are calculated using the current parameters, and the difference between the predicted value and the true value is calculated according to the loss function. Then, back propagation is performed to calculate the gradient of the loss function with respect to the model parameters and update the model parameters.

[0133] Algorithm model deployment: Deploy the trained model to the application of defect detection in the outer semi-conductive layer of high-voltage cables.

[0134] In combination with the above-mentioned contents, in a specific implementation, the process of performing feature recognition and extraction combining depthwise separable convolution with multi-scale feature fusion on the reconstructed detection image to obtain key image features may include the following sub-steps S21 to S25:

[0135] Step S21: inputting the reconstructed detection image into a pre-trained anomaly detection model;

[0136] Step S22: According to the anomaly detection model, combined with the edge detection algorithm, identify and extract the preliminary detection feature map of the reconstructed detection image;

[0137] Step S23: performing a depth-separable convolution on the preliminary detection feature map to obtain a depth-convolution feature map;

[0138] Step S24: inputting the deep convolution feature maps into a plurality of first convolution layers of different scales for convolution processing respectively, obtaining convolution scale feature maps corresponding to each first convolution layer, and fusing each convolution scale feature map with each other in a top-down and bottom-up manner to achieve multi-scale feature fusion, and obtaining a multi-scale fused feature map;

[0139] Step S25: Perform residual convolution processing on the multi-scale fusion feature map to obtain key image features.

[0140] Step 104 , performing anomaly detection based on defect type and position on the key image features, and outputting defect detection results of the outer semi-conductive layer of the high-voltage cable.

[0141] Furthermore, the specific process of performing abnormal detection on the key image features based on the defect category and position and outputting the defect detection result of the outer semi-conductive layer of the high-voltage cable can be achieved by executing the following sub-steps S31 to S34:

[0142] Step S31: performing anomaly detection based on defect category and location on key image features;

[0143] Step S32: when at least one abnormal feature is detected, the defect category closest to each abnormal feature is determined respectively, and a defect optimization detection frame of a corresponding size is generated at the location of the defect category; wherein the defect optimization detection frame is obtained by optimizing defect category recognition and position positioning based on a multi-task loss function during the training process of the abnormality detection model;

[0144] Step S33: performing feature comparison on the abnormal feature information corresponding to each defect detection frame and the preset defect feature information to determine whether an abnormal defect occurs in the outer semi-conductive layer of the high-voltage cable;

[0145] Step S34: Based on the feature comparison result, output the defect detection result of the outer semi-conductive layer of the high-voltage cable.

[0146] As an optional embodiment, a real-time alarm mechanism can be set to trigger an alarm in real time when a defect abnormality is detected in the outer semi-conductive layer of the high-voltage cable, and the relevant information of the abnormality detection can be sent to the management end of the high-voltage cable outer semi-conductive layer cutting equipment.

[0147] After outputting the defect detection result of the outer semi-conductive layer of the high-voltage cable, when the defect detection result indicates that an abnormal defect occurs in the outer semi-conductive layer of the high-voltage cable, a real-time alarm is triggered for the abnormal defect based on the real-time alarm mechanism, and the defect detection result is sent to the management end of the cutting equipment corresponding to the outer semi-conductive layer of the high-voltage cable.

[0148] Specifically, when a defect in the outer semi-conductive layer of the high-voltage cable is detected, the alarm mechanism is triggered in real time, and the maintenance personnel are notified by sound, vibration or other means; the defect detection results of the outer semi-conductive layer of the high-voltage cable are output, including the location, size, shape and other characteristics of the defect, as well as the possible defect type and level; based on the defect detection results, optional suggestions or action guidelines are provided to guide maintenance personnel to take appropriate measures; the alarm information and abnormal detection results are recorded for subsequent analysis and auditing.

[0149] More specifically, first, maintenance personnel need to analyze the defect detection results output after anomaly detection, including identifying the defect type, defect location, and defect severity.

[0150] Secondly, implement corresponding action guidelines based on the analysis and test results. Specifically:

[0151] When the defect level is minor, the defect location and type are recorded and the location is monitored regularly. When the defect level changes from minor to the next severity level (such as medium or severe), the maintenance personnel are notified. Therefore, for defects that have little impact on equipment operation, regular monitoring is adopted, which not only allows real-time attention to the defect situation at the location, but also saves personnel inspection costs.

[0152] When the defect level is medium, the maintenance personnel are notified to conduct a preliminary inspection and update the defect database to record the defect location, type and severity.

[0153] When the defect level is serious, the cable operation is immediately stopped, and the relevant personnel are notified through the alarm system to carry out emergency repairs and record the status before and after the repairs.

[0154] Finally, by combining the historical database and real-time detection results, the potential defect development trend is predicted through machine learning algorithms to achieve intelligent predictive maintenance. At the same time, the abnormal detection results are visualized, the defect markers are displayed in the defect detection software, and a real-time monitoring platform is established so that maintenance personnel can view the detection status in real time.

[0155] It should be noted that, in order to enable those skilled in the art to better distinguish data of the same type but with different actual meanings, some technical features are distinguished by the first, second, and third in the embodiments of the present invention. The first, second, and third are only used for data distinction and have no other special meanings. It can be understood that the present invention is not limited to this.

[0156] In an embodiment of the present invention, a method for detecting defects in the outer semi-conductive layer of a high-voltage cable based on machine vision is provided. (1) First, by collecting images of the outer semi-conductive layer of the high-voltage cable to be inspected, and based on image preprocessing and feature extraction, the clarity of the original image is significantly improved, the features of interest are enhanced, and electronic noise and optical interference are reduced, which is conducive to the detection and identification of the outer semi-conductive layer of the high-voltage cable. (2) Secondly, an improved adaptive convolutional network is adopted, and by combining deep separable convolution, multi-scale feature fusion processing and hierarchical feature learning, the key image features of the outer semi-conductive layer of the high-voltage cable are automatically learned, the detection accuracy of the model for defects is improved, and errors caused by human intervention and subjective judgment are avoided. (3) At the same time, the optimization loss function is introduced to perform abnormal detection on the outer semi-conductive layer of the high-voltage cable to improve the performance of the model, thereby more accurately identifying the defects of the outer semi-conductive layer of the high-voltage cable, adapting to image detection of different sizes and angles, and avoiding the situation of missing or misdetecting small target defects, further improving the accuracy and stability of detection. (4) Through the real-time alarm mechanism, a real-time alarm is triggered based on the abnormal defect detection results, and the relevant detection results are sent to the management end of the high-voltage cable outer semi-conductive layer cutting equipment to achieve real-time monitoring of the effect of the high-voltage cable outer semi-conductive layer cutting equipment, so as to timely discover and effectively deal with abnormal conditions and improve the reliability and stability of the cutting equipment.

[0157] In general, the technical solution of the present invention has the advantages of high automation, high detection accuracy, good real-time performance, etc., and can effectively improve the efficiency and accuracy of defect detection of the outer semi-conductive layer of high-voltage cables. By adopting this solution, the defect detection of the outer semi-conductive layer of high-voltage cables can be made more intelligent, efficient and accurate. This provides a reliable quality control method for the cutting equipment of the outer semi-conductive layer of high-voltage cables, and improves the overall operation effect of the cutting equipment.

[0158] For better explanation, refer to Figure 4, showing a schematic diagram of the overall process of a method for detecting defects in the outer semi-conductive layer of a high-voltage cable provided by an embodiment of the present invention. It should be pointed out that this embodiment only briefly describes the general process of detecting defects in the outer semi-conductive layer of a high-voltage cable, and the specific implementation process of each step can be understood by referring to the relevant content in the aforementioned embodiment, which will not be described here. It can be understood that the present invention does not limit this.

[0159] Step 401: collecting multiple defect images of different defects of the outer semi-conductive layer of the high-voltage cable, performing data preprocessing on the multiple defect images, and performing image reconstruction to obtain a reconstructed defect image data set;

[0160] Step 402: using the adaptive convolutional network including the auxiliary classifier as the network architecture, constructing an initial anomaly detection model, and adjusting parameters of the initial anomaly detection model;

[0161] Step 403: Based on the defined multi-task loss function, the initial anomaly detection model is trained and verified by reconstructing the defect image dataset;

[0162] Step 404: In the process of model training and verification, based on the model verification and evaluation results, the parameters of the multi-task loss function are iteratively optimized and updated using the gradient descent method to obtain an anomaly detection model;

[0163] Step 405: acquiring an image to be detected of the outer semi-conductive layer of the high-voltage cable, and reconstructing the image to be detected to obtain a reconstructed detection image;

[0164] Step 406: input the reconstructed detection image into the anomaly detection model, first perform feature recognition and extraction on the reconstructed detection image by combining depthwise separable convolution with multi-scale feature fusion, then perform anomaly detection based on defect category and position, and output the defect detection result of the outer semi-conductive layer of the high-voltage cable;

[0165] Step 407: When the defect detection result indicates that an abnormal defect occurs in the outer semi-conductive layer of the high-voltage cable, a real-time alarm is triggered for the abnormal defect based on the real-time alarm mechanism, and the defect detection result is sent to the management end of the cutting equipment corresponding to the outer semi-conductive layer of the high-voltage cable.

[0166] Reference Figure 5 , shows a structural block diagram of a high-voltage cable outer semi-conductive layer defect detection device provided by an embodiment of the present invention, which may specifically include:

[0167] A data acquisition unit 501 is used to acquire an image to be detected of the outer semi-conductive layer of the high-voltage cable;

[0168] An image reconstruction unit 502 is used to reconstruct the image to be detected to obtain a reconstructed detection image;

[0169] A feature extraction unit 503 is used to perform feature recognition and extraction on the reconstructed detection image by combining depthwise separable convolution with multi-scale feature fusion to obtain key image features;

[0170] The anomaly detection unit 504 is used to perform anomaly detection on the key image features based on defect type and position, and output defect detection results of the outer semi-conductive layer of the high-voltage cable.

[0171] In an optional embodiment, the image reconstruction unit 502 includes:

[0172] An image graying unit, used for converting the image to be detected into a grayscale image based on image graying;

[0173] A noise removal unit, used for removing image noise of the grayscale image by using a Gaussian blur filter to obtain a denoised image;

[0174] The image binarization unit is used to convert the denoised image into a binary image by image binarization, and separate the target area of ​​the binary image by threshold segmentation to obtain a reconstructed detection image.

[0175] In an optional embodiment, the feature extraction unit 503 includes:

[0176] A data input unit, used for inputting the reconstructed detection image into a pre-trained anomaly detection model;

[0177] A preliminary feature extraction unit, used to identify and extract a preliminary detection feature map of the reconstructed detection image according to the anomaly detection model in combination with an edge detection algorithm;

[0178] A depthwise separable convolution unit, used for performing a depthwise separable convolution on the preliminary detection feature map to obtain a depthwise convolution feature map;

[0179] A multi-scale feature fusion unit, used to input the depth convolution feature map into a plurality of first convolution layers of different scales for convolution processing, obtain convolution scale feature maps corresponding to each of the first convolution layers, and fuse the convolution scale feature maps with each other in a top-down and bottom-up manner to achieve multi-scale feature fusion and obtain a multi-scale fused feature map;

[0180] The residual convolution processing module is used to perform residual convolution processing on the multi-scale fusion feature map to obtain key image features.

[0181] In an optional embodiment, the abnormality detection unit 504 includes:

[0182] An anomaly detection subunit, used for performing anomaly detection on the key image features based on defect category and position;

[0183] A defect optimization detection frame generation unit is used to determine the defect category closest to each abnormal feature when at least one abnormal feature is detected, and to generate a defect optimization detection frame of a corresponding size at the location of the defect category, wherein the defect optimization detection frame is obtained by optimizing defect category recognition and position positioning based on a multi-task loss function during the training of the abnormality detection model;

[0184] A feature comparison unit, used for comparing the abnormal feature information corresponding to each defect detection frame with the preset defect feature information, so as to determine whether an abnormal defect occurs in the outer semi-conductive layer of the high-voltage cable;

[0185] The result output unit is used to output the defect detection result of the outer semi-conductive layer of the high-voltage cable based on the feature comparison result.

[0186] In an optional embodiment, the device further includes:

[0187] A defect image acquisition unit, used for acquiring multiple defect images of different defects of the outer semi-conductive layer of the high-voltage cable;

[0188] A data preprocessing unit, used to perform data preprocessing on the plurality of defect images and perform image reconstruction to obtain a reconstructed defect image data set;

[0189] A data set division unit, used for dividing the reconstructed defect image data set into an image defect training set and an image defect verification set according to a preset ratio;

[0190] The initial anomaly detection model construction unit is used to construct an initial anomaly detection model using an adaptive convolutional network including an auxiliary classifier as a network architecture; the initial anomaly detection model includes a feature extraction module, a multi-scale feature fusion module and a defect detection module; wherein:

[0191] The multi-scale feature fusion module adopts a feature pyramid network, and the feature pyramid network includes multiple first convolutional layers, and the multiple first convolutional layers are used to generate feature maps of different scales;

[0192] The feature extraction module includes at least one second convolutional layer and multiple third convolutional layers, the at least one second convolutional layer is used to perform preliminary feature extraction on the image of the input model, and each of the third convolutional layers adopts a convolution type of depthwise separable convolution;

[0193] The defect detection module is composed of a defect detection submodule and a feature comparison submodule. The defect detection submodule is used to perform anomaly detection on the key image features output by the multi-scale feature fusion module, and generate at least one defect detection frame based on the feature anomaly detection result, and the defect detection frame corresponds to the candidate detection area; the feature comparison submodule is used to compare and verify the abnormal feature information in the candidate detection area, and generate a defect detection result;

[0194] A parameter adjustment unit, configured to perform parameter adjustment on each of the first convolution layer, the second convolution layer, and the third convolution layer, wherein the parameter adjustment includes setting a convolution kernel size, a step size, and padding;

[0195] A model training and verification unit, configured to perform model training on the initial anomaly detection model through the image defect training set based on a defined multi-task loss function, and perform model verification on the trained anomaly detection model through the image defect verification set;

[0196] A loss function optimization unit, used to iteratively optimize and update the parameters of the multi-task loss function using a gradient descent method based on the model verification evaluation results during the model training verification process, so as to optimize the defect detection frame's ability to identify defect categories and locate defect locations;

[0197] The anomaly detection model output unit is used to stop iteration and output the final anomaly detection model when the number of iterations reaches a preset maximum number of iterations, or the model prediction error is less than or equal to a preset error threshold.

[0198] In an optional embodiment, the multi-task loss function is formed by comprehensively considering the position error, position weighting, classification error and category weighting of the defect detection frame; the position weighting is used to evaluate the shape and position change of the defect detection frame to reflect the importance of position positioning; the category weighting is used to evaluate the defect category of the abnormal features in the defect detection frame to reflect the importance of category identification; the specific expression of the multi-task loss function is as follows:

[0199]

[0200] in, represents the multi-task loss function, representing the total loss; and Represent the total weight of category weighting and position weighting respectively; and denote the category-weighted and position-weighted losses respectively.

[0201] In an optional embodiment, the device further includes:

[0202] A real-time alarm triggering unit is used to trigger a real-time alarm for the abnormal defect based on the real-time alarm mechanism when the defect detection result indicates that an abnormal defect has occurred in the outer semi-conductive layer of the high-voltage cable, and send the defect detection result to the management end of the cutting equipment corresponding to the outer semi-conductive layer of the high-voltage cable.

[0203] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the aforementioned method embodiment.

[0204] An embodiment of the present invention further provides an electronic device, the device comprising a processor and a memory:

[0205] The memory is used to store program codes and transmit the program codes to the processor;

[0206] The processor is used to execute the method for detecting defects in the outer semi-conductive layer of a high-voltage cable according to any embodiment of the present invention according to the instructions in the program code.

[0207] An embodiment of the present invention further provides a computer-readable storage medium, which is used to store program codes, and the program codes are used to execute the method for detecting defects in the outer semi-conductive layer of a high-voltage cable according to any embodiment of the present invention.

[0208] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0209] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0210] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0211] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0212] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0213] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting defects in the outer semi-conductive layer of a high-voltage cable, characterized in that: include: Acquire the image to be inspected of the outer semi-conductive layer of the high-voltage cable; Reconstructing the image to be detected to obtain a reconstructed detection image; Performing feature recognition and extraction combining depth-separable convolution with multi-scale feature fusion on the reconstructed detection image to obtain key image features; The key image features are subjected to abnormality detection based on defect categories and positions, and defect detection results of the outer semi-conductive layer of the high-voltage cable are output.

2. The method for detecting defects in the outer semi-conductive layer of a high-voltage cable according to claim 1, characterized in that: The step of reconstructing the image to be detected to obtain a reconstructed detection image includes: Based on image graying, converting the image to be detected into a grayscale image; Removing image noise from the grayscale image by using a Gaussian blur filter to obtain a denoised image; The denoised image is converted into a binary image by image binarization, and a target area of ​​the binary image is separated by threshold segmentation to obtain a reconstructed detection image.

3. The method for detecting defects in the outer semi-conductive layer of a high-voltage cable according to claim 1, characterized in that: The step of performing feature recognition and extraction combining deep separable convolution with multi-scale feature fusion on the reconstructed detection image to obtain key image features includes: Inputting the reconstructed detection image into a pre-trained anomaly detection model; According to the anomaly detection model, combined with an edge detection algorithm, identifying and extracting a preliminary detection feature map of the reconstructed detection image; Performing depth-separable convolution on the preliminary detection feature map to obtain a depth-convolution feature map; Inputting the depth convolution feature map into a plurality of first convolution layers of different scales for convolution processing respectively, obtaining convolution scale feature maps corresponding to each of the first convolution layers, and fusing the convolution scale feature maps with each other in a top-down and bottom-up manner to achieve multi-scale feature fusion, and obtaining a multi-scale fused feature map; Perform residual convolution processing on the multi-scale fusion feature map to obtain key image features.

4. The method for detecting defects in the outer semi-conductive layer of a high-voltage cable according to claim 3, characterized in that: The performing abnormality detection based on defect category and position on the key image features and outputting defect detection results of the outer semi-conductive layer of the high-voltage cable includes: Performing anomaly detection based on defect category and location on the key image features; When at least one abnormal feature is detected, the defect category closest to each abnormal feature is determined respectively, and a defect optimization detection frame of a corresponding size is generated at the location of the defect category. The defect optimization detection frame is obtained by optimizing defect category recognition and position positioning based on a multi-task loss function during the training of the abnormality detection model; Comparing the abnormal feature information corresponding to each defect detection frame with the preset defect feature information respectively, so as to determine whether an abnormal defect occurs in the outer semi-conductive layer of the high-voltage cable; Based on the feature comparison result, the defect detection result of the outer semi-conductive layer of the high-voltage cable is output.

5. The method for detecting defects in the outer semi-conductive layer of a high-voltage cable according to claim 4, characterized in that: The training process of the anomaly detection model includes: Collect multiple defect images of different defects in the outer semi-conductive layer of high-voltage cables; Performing data preprocessing on the plurality of defect images and performing image reconstruction to obtain a reconstructed defect image data set; Dividing the reconstructed defect image data set into an image defect training set and an image defect verification set according to a preset ratio; An initial anomaly detection model is constructed using an adaptive convolutional network including an auxiliary classifier as the network architecture; the initial anomaly detection model includes a feature extraction module, a multi-scale feature fusion module and a defect detection module; wherein: The multi-scale feature fusion module adopts a feature pyramid network, and the feature pyramid network includes multiple first convolutional layers, and the multiple first convolutional layers are used to generate feature maps of different scales; The feature extraction module includes at least one second convolutional layer and multiple third convolutional layers, the at least one second convolutional layer is used to perform preliminary feature extraction on the image of the input model, and each of the third convolutional layers adopts a convolution type of depthwise separable convolution; The defect detection module is composed of a defect detection submodule and a feature comparison submodule. The defect detection submodule is used to perform anomaly detection on the key image features output by the multi-scale feature fusion module, and generate at least one defect detection frame based on the feature anomaly detection result, and the defect detection frame corresponds to the candidate detection area; the feature comparison submodule is used to compare and verify the abnormal feature information in the candidate detection area, and generate a defect detection result; Performing parameter adjustment on each of the first convolutional layer, the second convolutional layer, and the third convolutional layer, wherein the parameter adjustment includes setting a convolution kernel size, a step size, and padding; Based on the defined multi-task loss function, the initial anomaly detection model is trained by the image defect training set, and the trained anomaly detection model is verified by the image defect verification set; In the process of model training and verification, based on the model verification and evaluation results, the gradient descent method is used to iteratively optimize and update the parameters of the multi-task loss function to optimize the defect detection box's ability to identify defect categories and locate defect locations; When the number of iterations reaches the preset maximum number of iterations, or the model prediction error is less than or equal to the preset error threshold, the iteration is stopped and the final anomaly detection model is output.

6. The method for detecting defects in the outer semi-conductive layer of a high-voltage cable according to claim 5, characterized in that: The multi-task loss function is formed by comprehensively considering the position error, position weighting, classification error and category weighting of the defect detection frame; the position weighting is used to evaluate the shape and position change of the defect detection frame to reflect the importance of position positioning; the category weighting is used to evaluate the defect category of the abnormal features in the defect detection frame to reflect the importance of category identification; the specific expression of the multi-task loss function is as follows: in, represents the multi-task loss function, representing the total loss; and Represent the total weight of category weighting and position weighting respectively; and denote the category-weighted and position-weighted losses respectively.

7. The method for detecting defects in the outer semi-conductive layer of a high-voltage cable according to any one of claims 1 to 6, characterized in that: After outputting the defect detection result of the outer semiconductive layer of the high-voltage cable, the method further comprises: When the defect detection result indicates that an abnormal defect occurs in the outer semi-conductive layer of the high-voltage cable, a real-time alarm is triggered for the abnormal defect based on a real-time alarm mechanism, and the defect detection result is sent to the management end of the cutting equipment corresponding to the outer semi-conductive layer of the high-voltage cable.

8. A device for detecting defects in the outer semi-conductive layer of a high-voltage cable, characterized in that: include: A data acquisition unit, used for acquiring an image to be detected of the outer semi-conductive layer of the high-voltage cable; An image reconstruction unit, used for reconstructing the image to be detected to obtain a reconstructed detection image; A feature extraction unit, used to perform feature recognition and extraction on the reconstructed detection image by combining deep separable convolution with multi-scale feature fusion to obtain key image features; The anomaly detection unit is used to perform anomaly detection on the key image features based on defect type and position, and output defect detection results of the outer semi-conductive layer of the high-voltage cable.

9. An electronic device, characterized in that: The device comprises a processor and a memory: The memory is used to store program codes and transmit the program codes to the processor; The processor is used to execute the method for detecting defects in the outer semiconductive layer of a high-voltage cable according to any one of claims 1 to 7 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program codes, and the program codes are used to execute the method for detecting defects in the outer semiconductive layer of a high-voltage cable according to any one of claims 1 to 7.

Images