Glass insulator spontaneous explosion defect identification method and device, computer equipment and storage medium
The self-destruction defect recognition model completed by the training processed the inspection image of the glass insulator, which solves the problem of low self-destruction defect detection accuracy in the prior art, and achieves higher self-destruction defect recognition accuracy.
Patent Information
- Application Number
- CN202411579567.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-07
- Publication Date
- 2025-02-07
AI Technical Summary
The prior art has low accuracy when detecting self-destruction defects of glass insulators, especially since the abnormal gaps generated by self-destruction are not obvious, resulting in low accuracy in detecting self-destruction defects of glass insulators in partial order.
Provide a method for self-destruction defect recognition of glass insulators. By obtaining the self-destruction defect recognition model completed by the training, the model trains the initial self-destruction defect recognition model through the abnormal image samples corresponding to the abnormal category and the normal image samples corresponding to the normal category to enhance its Global information processing capability. Then, the self-destruction defect identification process is performed on the inspection images of multiple consecutive glass insulators through this model, and the self-destruction defect identification results of each glass insulator are obtained.
It effectively improves the accuracy of self-destruction defect identification of glass insulators, and can more accurately identify self-destruction defects of glass insulators in different orders, reducing false detection and missed detection.
Smart Images
Figure CN119399744A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of power transmission lines, and in particular to a method, device, computer equipment, storage medium and computer program product for identifying self-explosion defects of glass insulators. Background Art
[0002] Glass insulators in transmission lines are one of the important components of the power grid in the process of transporting electricity. Their main function is to support the transmission wires, isolate the current, and protect the power components on the transmission lines. However, due to the quality of the material itself, environmental factors, and the influence of electrical stress, the balance of compressive stress and tensile stress in the glass parts of the glass insulators will be destroyed, and cracks will quickly appear under the action of stress, causing the glass parts to shatter, that is, self-explosion. The protective effect of the glass insulator string that explodes will be greatly reduced, which is an important safety hazard on the transmission line.
[0003] In order to detect whether a glass insulator will have abnormal defects such as self-explosion, the existing technology often uses deep learning or machine learning algorithms to detect the abnormal gap of the glass insulator. However, the abnormal gap caused by the self-explosion of some orders of glass insulators is not obvious, resulting in low detection accuracy of the self-explosion defects of these glass insulators. Summary of the invention
[0004] Based on this, it is necessary to provide a glass insulator self-explosion defect identification method, device, computer equipment, computer-readable storage medium and computer program product that can improve the detection accuracy of glass insulator self-explosion defects in response to the above technical problems.
[0005] In a first aspect, the present application provides a method for identifying a self-explosion defect of a glass insulator. The method comprises:
[0006] Acquire a trained self-detonation defect recognition model; the trained self-detonation defect recognition model is obtained by training an initial self-detonation defect recognition model through abnormal image samples corresponding to abnormal categories and normal image samples corresponding to normal categories; the abnormal categories are obtained according to abnormal feature settings corresponding to glass insulators of different orders; the initial self-detonation defect recognition model is obtained by performing global information enhancement processing on the original self-detonation defect recognition model;
[0007] Obtain inspection images of multiple consecutive glass insulators;
[0008] The self-explosion defect recognition model that has been trained is used to perform self-explosion defect recognition processing on the inspection image to obtain a self-explosion defect recognition result for each glass insulator.
[0009] In one embodiment, obtaining a trained self-explosion defect recognition model includes:
[0010] Acquire abnormal image samples of continuous glass insulators corresponding to the abnormal category and normal image samples of continuous glass insulators corresponding to the normal category;
[0011] Inputting the abnormal image sample and the normal image sample into the initial self-explosion defect recognition model to obtain the predicted category corresponding to the abnormal image sample and the predicted category corresponding to the normal image sample;
[0012] According to the category difference between the predicted category corresponding to the abnormal image sample and the abnormal category corresponding to the abnormal image sample, and the category difference between the predicted category corresponding to the normal image sample and the normal category, the initial self-destruction defect recognition model is iteratively trained to obtain the trained self-destruction defect recognition model.
[0013] In one embodiment, obtaining abnormal image samples corresponding to the abnormal category of continuous glass insulators and normal image samples corresponding to the normal category of continuous glass insulators includes:
[0014] Acquiring abnormal images corresponding to the glass insulators of different orders, and acquiring normal images corresponding to the glass insulators of different orders;
[0015] Performing feature extraction processing on the abnormal image to obtain steel cap group features of the abnormal image corresponding to the first glass insulator, abnormal gap features of the abnormal image corresponding to the middle glass insulator, and connection fittings and tail fittings features of the abnormal image corresponding to the last glass insulator;
[0016] Determine the abnormality category of the glass insulator according to the characteristics of the steel cap group, the characteristics of the abnormal gap, and the characteristics of the connecting hardware and the tail hardware;
[0017] According to the abnormal category, the abnormal image is labeled to obtain abnormal image samples corresponding to the abnormal category, and the normal image is labeled as a normal category to obtain normal image samples corresponding to the normal category.
[0018] In one embodiment, before inputting the abnormal image sample and the normal image sample into the initial self-explosion defect recognition model to obtain the predicted category corresponding to the abnormal image sample and the predicted category corresponding to the normal image sample, the method further includes:
[0019] According to the convolutional network backbone and the feature fusion unit, the original self-explosion defect recognition model is obtained;
[0020] The multi-head self-attention network and the original self-destruction defect recognition model are fused to obtain the initial self-destruction defect recognition model.
[0021] In one embodiment, the self-explosion defect recognition model that has been trained is used to perform self-explosion defect recognition processing on the inspection image to obtain a self-explosion defect recognition result for each glass insulator, including:
[0022] Through the convolutional network backbone in the self-explosion defect recognition model, the inspection image is subjected to feature extraction processing and global response normalization processing based on self-supervised learning to obtain feature data of the inspection image;
[0023] Performing global information analysis and processing on the feature data through the multi-head self-attention network in the self-explosion defect recognition model to obtain updated feature data of the inspection image;
[0024] The updated feature data is subjected to feature fusion processing by a feature fusion unit in the self-explosion defect recognition model to obtain a category recognition result of the inspection image;
[0025] According to the category recognition result of the inspection image corresponding to each glass insulator, the self-explosion defect recognition result of each glass insulator is obtained.
[0026] In one embodiment, the convolutional network backbone in the self-explosion defect recognition model is used to perform feature extraction processing and global response normalization processing based on self-supervised learning on the inspection image to obtain feature data of the inspection image, and further includes:
[0027] Performing mask processing on the inspection image through the convolutional network backbone to obtain a mask image of the inspection image;
[0028] Through the encoder and decoder of the convolutional network backbone, the mask image is subjected to mask prediction processing to obtain a predicted image corresponding to the mask image; the predicted image is used to represent the predicted data of the mask area image in the mask image;
[0029] The predicted image, the mask image and the inspection image are sequentially subjected to global feature aggregation processing, feature normalization processing and feature calibration processing to obtain the feature data.
[0030] In a second aspect, the present application also provides a device for identifying self-explosion defects of glass insulators. The device comprises:
[0031] The model acquisition module is used to acquire a trained self-detonation defect recognition model; the trained self-detonation defect recognition model is obtained by training an initial self-detonation defect recognition model through abnormal image samples corresponding to abnormal categories and normal image samples corresponding to normal categories; the abnormal categories are obtained according to abnormal feature settings corresponding to glass insulators of different orders; the initial self-detonation defect recognition model is obtained by performing global information enhancement processing on the original self-detonation defect recognition model;
[0032] An image acquisition module is used to acquire inspection images of multiple consecutive glass insulators;
[0033] The defect recognition module is used to perform self-explosion defect recognition processing on the inspection image through the trained self-explosion defect recognition model to obtain the self-explosion defect recognition result of each glass insulator.
[0034] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:
[0035] Acquire a trained self-detonation defect recognition model; the trained self-detonation defect recognition model is obtained by training an initial self-detonation defect recognition model through abnormal image samples corresponding to abnormal categories and normal image samples corresponding to normal categories; the abnormal categories are obtained according to abnormal feature settings corresponding to glass insulators of different orders; the initial self-detonation defect recognition model is obtained by performing global information enhancement processing on the original self-detonation defect recognition model;
[0036] Obtain inspection images of multiple consecutive glass insulators;
[0037] The self-explosion defect recognition model that has been trained is used to perform self-explosion defect recognition processing on the inspection image to obtain a self-explosion defect recognition result for each glass insulator.
[0038] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:
[0039] Acquire a trained self-detonation defect recognition model; the trained self-detonation defect recognition model is obtained by training an initial self-detonation defect recognition model through abnormal image samples corresponding to abnormal categories and normal image samples corresponding to normal categories; the abnormal categories are obtained according to abnormal feature settings corresponding to glass insulators of different orders; the initial self-detonation defect recognition model is obtained by performing global information enhancement processing on the original self-detonation defect recognition model;
[0040] Obtain inspection images of multiple consecutive glass insulators;
[0041] The self-explosion defect recognition model that has been trained is used to perform self-explosion defect recognition processing on the inspection image to obtain a self-explosion defect recognition result for each glass insulator.
[0042] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:
[0043] Acquire a trained self-detonation defect recognition model; the trained self-detonation defect recognition model is obtained by training an initial self-detonation defect recognition model through abnormal image samples corresponding to abnormal categories and normal image samples corresponding to normal categories; the abnormal categories are obtained according to abnormal feature settings corresponding to glass insulators of different orders; the initial self-detonation defect recognition model is obtained by performing global information enhancement processing on the original self-detonation defect recognition model;
[0044] Obtain inspection images of multiple consecutive glass insulators;
[0045] The self-explosion defect recognition model that has been trained is used to perform self-explosion defect recognition processing on the inspection image to obtain a self-explosion defect recognition result for each glass insulator.
[0046] The above-mentioned glass insulator self-explosion defect recognition method, device, computer equipment, storage medium and computer program product obtain a trained self-explosion defect recognition model; the trained self-explosion defect recognition model is obtained by training the initial self-explosion defect recognition model through abnormal image samples corresponding to abnormal categories and normal image samples corresponding to normal categories; the abnormal categories are obtained according to abnormal feature settings corresponding to glass insulators of different orders; the initial self-explosion defect recognition model is obtained by performing global information enhancement processing on the original self-explosion defect recognition model; the inspection images of multiple consecutive glass insulators are obtained; the inspection images are processed by self-explosion defect recognition through the trained self-explosion defect recognition model to obtain the self-explosion defect recognition result of each glass insulator. The method is adopted to effectively improve the self-explosion defect recognition ability of the enhanced self-explosion defect recognition model by enhancing the global information processing ability of the initial self-explosion defect recognition model, and the self-explosion defect conditions of glass insulators of different orders can be more accurately identified by collecting the inspection images of multiple consecutive glass insulators for self-explosion defect recognition processing, thereby further improving the accuracy of self-explosion defect recognition of glass insulators. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 A diagram showing an application environment of a method for identifying a self-explosion defect of a glass insulator in an embodiment;
[0048] Figure 2A schematic diagram of a process of identifying a self-explosion defect of a glass insulator in one embodiment;
[0049] Figure 3 A schematic flow chart of the steps of obtaining abnormal image samples corresponding to continuous glass insulators in abnormal categories and normal image samples corresponding to continuous glass insulators in normal categories in one embodiment;
[0050] Figure 4 The figure is a flow chart of the steps of performing self-destruction defect recognition processing on the inspection image by the self-destruction defect recognition model completed through training in one embodiment;
[0051] Figure 5 A structural schematic diagram of a self-destruction defect recognition model completed through training in an embodiment, and steps of performing self-destruction defect recognition processing on an inspection image;
[0052] Figure 6 Schematic diagram of the network structure of a multi-head self-attention network in one embodiment;
[0053] Figure 7 A schematic diagram of a network structure of a feature fusion unit in an embodiment;
[0054] Figure 8 Schematic diagram of the principle of a self-supervised learning framework in one embodiment;
[0055] Fig. 9 A schematic diagram of a flow chart of a method for identifying a self-explosion defect of a glass insulator in another embodiment;
[0056] Fig.10 A schematic diagram of a flow chart of a method for identifying a self-explosion defect of a glass insulator in another embodiment;
[0057] Fig.11 is a structural block diagram of a device for identifying a self-explosion defect of a glass insulator in an embodiment;
[0058] Fig.12 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION
[0059] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0060] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0061] The method for identifying the self-explosion defect of a glass insulator provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Among them, the image acquisition device 101 communicates with the server 102 through the network. The data storage system can store the data that the server 102 needs to process. The data storage system can be integrated on the server 102, or it can be placed on the cloud or other network servers. The image acquisition device 101 is used to inspect glass insulators in the target scene (such as the transmission line scene in the power grid) and collect inspection images, and transmit the inspection images to the server 102 for processing. Among them, the image acquisition device 101 can be implemented by a drone or a drone cluster. The server 102 can be implemented by an independent server or a server cluster consisting of multiple servers.
[0062] In one embodiment, Figure 2 As shown in the figure, a method for identifying the self-explosion defect of a glass insulator is provided. Figure 1 The server in the example is used to illustrate the following steps:
[0063] Step S201, obtaining a trained self-detonation defect recognition model; the trained self-detonation defect recognition model is trained on an initial self-detonation defect recognition model by using abnormal image samples corresponding to abnormal categories and normal image samples corresponding to normal categories; the abnormal categories are set according to abnormal features corresponding to glass insulators in different orders; the initial self-detonation defect recognition model is obtained by performing global information enhancement processing on the original self-detonation defect recognition model.
[0064] Among them, the self-explosion defect recognition model is used to identify whether the glass insulator has self-explosion defects through the input image (such as the inspection image). The abnormal category refers to the defect category of the glass insulator that has self-explosion; the abnormal category includes defect categories with different self-explosion situations. The normal category refers to the category of the glass insulator that has not self-exploded.
[0065] The "order" in the glass insulators of different orders refers to the order of detecting each glass insulator in the continuous glass insulators.
[0066] Specifically, the server can use a network model that is relatively strong in global information processing in advance to perform global information enhancement processing on the original self-destruction defect recognition model to obtain an initial self-destruction defect recognition model to be trained. Then, according to the self-destruction defect characteristics of the glass insulator, the abnormal category of the glass insulator is determined; using the abnormal image samples corresponding to the annotated abnormal category and the normal image samples corresponding to the normal category, the initial self-destruction defect recognition model is iteratively trained to obtain a trained self-destruction defect recognition model.
[0067] Step S202, obtaining inspection images of a plurality of consecutive glass insulators.
[0068] Among them, the inspection image refers to the image captured by the image acquisition device of the glass insulator in the target scene (such as the transmission line scene in the power grid); the inspection image is used to analyze and identify whether the corresponding glass insulator has a self-explosion defect.
[0069] Specifically, the drone can inspect the glass insulators of the target transmission line and take multi-scale and multi-angle images of continuous glass insulators to obtain inspection images; the drone then transmits the collected inspection images of multiple continuous glass insulators to the server.
[0070] Step S203, using the trained self-explosion defect recognition model, performs self-explosion defect recognition processing on the inspection image to obtain a self-explosion defect recognition result for each glass insulator.
[0071] Specifically, the server inputs the inspection image into a trained self-detonation defect recognition model to identify whether there is a self-detonation defect in the inspection image through the self-detonation defect recognition model, and outputs a category recognition result of the inspection image in terms of the self-detonation defect; then the server determines the self-detonation defect recognition result of the glass insulator corresponding to the inspection image based on the category recognition result of the inspection image, and finally processes to obtain the self-detonation defect recognition result of each glass insulator.
[0072] In the above-mentioned glass insulator self-explosion defect recognition method, a trained self-explosion defect recognition model is obtained; the trained self-explosion defect recognition model is trained to obtain the initial self-explosion defect recognition model through abnormal image samples corresponding to abnormal categories and normal image samples corresponding to normal categories; the abnormal categories are obtained according to abnormal feature settings corresponding to glass insulators of different orders; the initial self-explosion defect recognition model is obtained by performing global information enhancement processing on the original self-explosion defect recognition model; inspection images of multiple consecutive glass insulators are obtained; the self-explosion defect recognition processing is performed on the inspection images through the trained self-explosion defect recognition model to obtain the self-explosion defect recognition result of each glass insulator. By adopting this method, the global information processing capability of the initial self-explosion defect recognition model is enhanced, and the self-explosion defect recognition capability of the enhanced self-explosion defect recognition model is effectively improved. By collecting inspection images of multiple consecutive glass insulators for self-explosion defect recognition processing, the self-explosion defect conditions of glass insulators of different orders can be more accurately identified, thereby further improving the accuracy of self-explosion defect recognition of glass insulators.
[0073] In one embodiment, the above step S201, obtaining a trained self-destruction defect recognition model, specifically includes the following contents: obtaining abnormal image samples corresponding to abnormal categories of continuous glass insulators and normal image samples corresponding to normal categories of continuous glass insulators; inputting the abnormal image samples and the normal image samples into the initial self-destruction defect recognition model to obtain the predicted categories corresponding to the abnormal image samples and the predicted categories corresponding to the normal image samples; according to the category difference between the predicted categories corresponding to the abnormal image samples and the abnormal categories corresponding to the abnormal image samples, and the category difference between the predicted categories corresponding to the normal image samples and the normal categories, iteratively training the initial self-destruction defect recognition model to obtain the trained self-destruction defect recognition model.
[0074] Specifically, the server collects abnormal images of glass insulators with self-explosion defects and normal images of glass insulators without self-explosion; by marking the abnormal categories corresponding to the abnormal images, abnormal image samples corresponding to the abnormal categories are obtained; by marking the normal categories corresponding to the normal images, normal image samples corresponding to the normal categories are obtained. The server divides the abnormal image samples and the normal image samples into a test set and a training set, and inputs the training set into the initial self-explosion defect recognition model to be trained, and the initial self-explosion defect recognition model outputs the predicted category corresponding to the training set; the server iteratively updates the model parameters of the initial self-explosion defect recognition model according to the category difference between the predicted category corresponding to the training set and the actual category (such as the normal category or the abnormal category) corresponding to the training set. When the preset training condition is reached, the server terminates the training and obtains the trained self-explosion defect recognition model. Among them, the preset training condition can be that the accuracy of the test set exceeds the preset threshold and / or the current number of iterations reaches the preset maximum number of iterations, etc.
[0075] In this embodiment, iterative training of the initial self-detonation defect recognition model is implemented, so that the trained self-detonation defect recognition model has the ability to recognize the self-detonation defects of glass insulators, and the accuracy of self-detonation defect recognition of glass insulators is improved with the help of the improved deep learning algorithm.
[0076] In one embodiment, Figure 3 As shown, obtaining abnormal image samples corresponding to the abnormal category of continuous glass insulators and normal image samples corresponding to the normal category of continuous glass insulators specifically includes the following contents:
[0077] Step S301, obtaining abnormal images corresponding to glass insulators of different orders, and obtaining normal images corresponding to glass insulators of different orders.
[0078] Specifically, for continuous glass insulators, the phenomena caused by the explosion of glass insulators in different orders are different. For example, the abnormal gap caused by the explosion of the first glass insulator is not obvious, while the abnormal gap caused by the explosion of the middle glass insulator is more obvious, and the abnormal gap caused by the explosion of the last glass insulator is not obvious, and there is no continuous insulator steel cap group after the explosion. Therefore, the server needs to obtain the abnormal images and normal images corresponding to each order of glass insulators to conduct in-depth analysis on the self-explosion defects of glass insulators in different orders.
[0079] Step S302, performing feature extraction processing on the abnormal image to obtain the steel cap group features of the abnormal image corresponding to the first glass insulator, the abnormal gap features of the abnormal image corresponding to the middle glass insulator, and the connection fittings and tail fittings features of the abnormal image corresponding to the last glass insulator.
[0080] Specifically, the server can extract the feature data of the self-explosion of glass insulators of different orders by performing feature extraction processing on the abnormal images of glass insulators of different orders: (1) Although the abnormal gap caused by the self-explosion of the first glass insulator is not obvious, two or more continuous insulator steel caps will be exposed after the self-explosion. This feature is relatively obvious and applicable to the self-explosion of the first glass insulator and the self-explosion of the glass insulator with the image acquisition angle facing the middle. The server obtains the steel cap group feature of the abnormal image corresponding to the first glass insulator and / or the glass insulator with the image acquisition angle facing the middle. (2) The self-explosion of the last glass insulator not only has an unclear abnormal gap but also no continuous insulator steel cap group after the self-explosion. However, the tail fitting structure is different from the insulator steel cap. It can form the unique bottom gap feature of the insulator steel cap connection fitting + tail fitting with the insulator steel cap connection fitting that remains after the self-explosion. This feature is significantly different from the feature of the continuous insulator steel cap group. The server obtains the connection fitting and tail fitting features of the abnormal image corresponding to the last glass insulator. (3) From a large-scale, long-distance perspective and a bird's-eye view, the abnormal gap caused by the self-explosion of the glass insulator in the middle piece still has relatively obvious characteristics, so the server obtains the abnormal gap characteristics of the abnormal image corresponding to the glass insulator in the middle piece.
[0081] Step S303, determining the abnormal category of the glass insulator according to the steel cap group characteristics, abnormal gap characteristics, and connection hardware and tail hardware characteristics.
[0082] Specifically, the server sets the glass insulator self-explosion steel cap group, glass insulator bottom self-explosion gap and glass insulator sub-abnormal gap as abnormal categories of glass insulators according to the characteristics of steel cap group, abnormal gap characteristics and connection hardware and tail hardware characteristics. In addition, due to the various types and shapes of hardware connecting the top of the insulator string, the connection hardware and the first insulator steel cap are easily misdetected as continuous insulator steel cap groups. Therefore, the first insulator steel cap is added as an auxiliary category, and the model's recognition ability of the insulator steel cap group is enhanced through inter-class variance, that is, the four categories of glass insulator top steel cap, glass insulator self-explosion steel cap group, glass insulator bottom self-explosion gap and glass insulator sub-abnormal gap are set as abnormal categories of glass insulators.
[0083] Step S304: annotate the abnormal images according to the abnormal categories to obtain abnormal image samples corresponding to the abnormal categories, and annotate the normal images as normal categories to obtain normal image samples corresponding to the normal categories.
[0084] Specifically, the server labels the abnormal images with their corresponding abnormal categories to obtain abnormal image samples, and labels the normal images with normal categories to obtain normal image samples.
[0085] In this embodiment, by analyzing the features of abnormal images corresponding to glass insulators of different orders, the abnormal category of the glass insulator is accurately determined, so that the self-explosion defect recognition model obtained by training with abnormal image samples and normal image samples in subsequent steps can identify the self-explosion of the first glass insulator by detecting continuous insulator steel cap groups, and identify the self-explosion of the last glass insulator by detecting the unique bottom gap between the insulator steel cap connecting hardware and the tail hardware, and identify the self-explosion of the middle glass insulator by detecting the abnormal gap, and accurately distinguish the steel cap of the first glass insulator and the continuous insulator steel cap group, thereby realizing composite cross-recognition of glass insulator self-explosion defects.
[0086] In one embodiment, before the abnormal image samples and the normal image samples are input into the initial self-destruction defect recognition model to obtain the predicted categories corresponding to the abnormal image samples and the predicted categories corresponding to the normal image samples, it also includes: obtaining the original self-destruction defect recognition model according to the convolutional network backbone and the feature fusion unit; fusing the multi-head self-attention network and the original self-destruction defect recognition model to obtain the initial self-destruction defect recognition model.
[0087] Specifically, the server can construct an original self-destruction defect recognition model based on a convolutional neural network, wherein the convolutional neural network includes a convolutional network backbone and a feature fusion unit. Then, a multi-head self-attention network with strong long-distance modeling capability is added to the original self-destruction defect recognition model to enhance the initial self-destruction defect recognition model's ability to aggregate global information and thus improve its detection capability.
[0088] In this embodiment, an initial self-explosion defect recognition model combining a convolutional neural network and a multi-head self-attention network is constructed, and the advantages brought by the structures of the convolutional neural network and the multi-head self-attention network are used to complement each other, thereby fundamentally improving the defect detection capability of the initial self-explosion defect recognition model and reducing false detections caused by foreground and background misalignment and overlap and the complex connection relationship of the production environment due to the self-explosion of glass insulators.
[0089] In one embodiment, Figure 4 As shown, in the above step S203, the self-explosion defect recognition model is trained to perform self-explosion defect recognition processing on the inspection image to obtain the self-explosion defect recognition result of each glass insulator, which specifically includes the following contents:
[0090] Step S401, through the convolutional network backbone in the self-explosion defect recognition model, the inspection image is subjected to feature extraction processing based on self-supervised learning and global response normalization processing to obtain feature data of the inspection image.
[0091] It should be noted that the self-explosion defect recognition model mainly includes three parts: convolutional network backbone, multi-head self-attention network and feature fusion unit. In practical applications, the convolutional network backbone can be implemented by ConvneXtv2 (ConvolutionalNext Version 2, an improved convolutional neural network architecture), the multi-head self-attention network can be implemented by DaViT (Dual Attention Vision Transformers, a transformer architecture optimized for visual tasks), and the feature fusion unit can be implemented by FPN (Feature Pyramid Network) + PAN (Path Aggregation Network). Figure 5 This is a structural diagram of the self-destruction defect recognition model that has been trained and the steps of performing self-destruction defect recognition processing on inspection images.
[0092] Specifically, the server performs feature extraction and global response normalization processing based on self-supervised learning on the inspection image through the convolutional network backbone; the traditional multi-head self-attention network DaViT requires a lot of computing resources. In order to reduce the amount of calculation, this embodiment improves the input data of the multi-head self-attention network DaViT, such as Figure 5 As shown in the figure, the feature data after downsampling of the stem layer in the convolutional network backbone is input into the multi-head self-attention network as data, and the convolutional network level features (s1-s4) and the multi-head attention network level features (s1-s4) are fused.
[0093] Step S402, performing global information analysis and processing on the feature data through the multi-head self-attention network in the self-explosion defect recognition model to obtain updated feature data of the inspection image.
[0094] Specifically, Figure 6 This is a schematic diagram of the network structure of the multi-head self-attention network. The multi-head self-attention network introduces a dual attention mechanism in global modeling. The innovation of the multi-head self-attention network is to perform self-attention from two orthogonal perspectives, modeling the spatial dimension and channel dimension respectively. In this way, the multi-head self-attention network can more efficiently capture the global information in the inspection image, thereby outputting updated feature data. Figure 6 In the above figure, P represents the total number of patches; C represents the total number of channels; N w Indicates the total number of windows (Number of windows); N hIndicates the number of detection heads (Number of heads); N g Indicates the number of channel groups (Number of channel groups); P w Indicates the number of patches per window (Patches per window); C h Indicates the number of detection heads per channel (Channels per head); C g Indicates the number of channels per group (Channelsper group); Represents matrix product (Matrix Product); ⊕ represents element-wise sum (Elementwise Sum).
[0095] Step S403, performing feature fusion processing on the updated feature data through the feature fusion unit in the self-explosion defect recognition model to obtain a category recognition result of the inspection image.
[0096] Specifically, Figure 7 The network structure diagram of the feature fusion unit. The feature fusion unit adopts the structure of FPN+PAN. FPN can transfer the deep semantic information in the updated feature data to the shallow features with higher resolution, which can make the large target clearer. PAN transfers the fine semantic information of the shallow features to the deep features, which makes the small targets clearer. Finally, the self-explosion defect recognition model outputs the category recognition result of the inspection image.
[0097] Step S404, obtaining a self-explosion defect recognition result of each glass insulator according to the category recognition result of the inspection image corresponding to each glass insulator.
[0098] Specifically, the server determines the self-explosion defect recognition result of each glass insulator by combining the category recognition results of one or more inspection images corresponding to each glass insulator. For example, as long as one corresponding inspection image of a glass insulator is marked as an abnormal category, the self-explosion defect recognition result of the glass insulator is confirmed to be the existence of a self-explosion defect; for another example, the number of category recognition results of one or more inspection images corresponding to each glass insulator is combined, and the category recognition results with the largest number are determined as the self-explosion defect recognition result of the glass insulator.
[0099] In this embodiment, the convolutional network backbone, multi-head self-attention network and feature fusion unit in the self-explosion defect recognition model cooperate with each other to realize the self-explosion defect recognition processing of the inspection image, effectively improving the accuracy of the self-explosion defect recognition results of each glass insulator obtained by processing, and solving the problem of false detection caused by the unclear abnormal gap of the first, last and other special angles of the glass insulators.
[0100] In one embodiment, the above step S401, through the convolutional network backbone in the self-explosion defect recognition model, performs feature extraction processing and global response normalization processing based on self-supervised learning on the inspection image to obtain feature data of the inspection image, specifically including the following contents: through the convolutional network backbone, the inspection image is masked to obtain a mask image of the inspection image; through the encoder and decoder of the convolutional network backbone, the mask image is masked to obtain a predicted image corresponding to the mask image; the predicted image is used to characterize the predicted data of the mask area image in the mask image; the predicted image, the mask image and the inspection image are sequentially subjected to global feature aggregation processing, feature normalization processing and feature calibration processing to obtain feature data.
[0101] It should be noted that, unlike traditional pure convolutional neural networks, the convolutional network backbone ConvneXtv2 contains the FCMAE self-supervised learning framework and the global response normalization (GRN) unit. The convolutional network backbone ConvneXtv2 significantly improves the performance of pure convolutional neural networks on recognition benchmarks through self-supervised learning technology and improved architecture.
[0102] Specifically, the inspection image is masked through the self-supervised learning framework in the convolutional network backbone to obtain a mask image of the inspection image; the mask image is then input into an encoder (such as a hierarchical encoder) and a decoder (such as a plain deorder) in the self-supervised learning framework to perform mask prediction processing on the mask area image in the mask image through the encoder and the decoder to obtain a predicted image corresponding to the mask image. Figure 8 Schematic diagram of the principle of self-supervised learning framework.
[0103] Furthermore, the spatial features of the predicted image, the mask image, and the inspection image are aggregated by the global response normalization unit to obtain the feature vector of the spatial feature, which can be obtained by a global The function stretches the spatial feature X (H*W*C) into a vector gx, where H, W, and C are the height, width, and number of channels of the X feature, respectively. The function can be expressed as the following formula:
[0104]
[0105] The feature vector is normalized by the global response normalization unit to scale the feature vector to the range of 0-1 to obtain normalized features. The feature normalization process can be expressed as the following formula:
[0106]
[0107] N(•) is the global response normalization function, is the L2 norm of the i-th channel. For the i-th channel, this formula calculates its importance relative to all other channels. The feature vector is subjected to feature calibration processing by the global response normalization unit, and the original feature data of the original input inspection image can be calibrated using the calculated normalized features to obtain feature data. The feature calibration process can be expressed as the following formula:
[0108]
[0109] In this embodiment, the inspection image is masked through the convolutional network backbone to obtain a mask image of the inspection image; the mask image is masked predicted through the encoder and decoder of the convolutional network backbone to obtain a predicted image corresponding to the mask image; feature data is obtained by sequentially performing global feature aggregation processing, feature normalization processing and feature calibration processing on the predicted image, mask image and inspection image, which solves the problem of low recognition accuracy of traditional convolutional networks and effectively improves the quality of the processed feature data.
[0110] In one embodiment, Fig. 9 As shown in the figure, another method for identifying the self-explosion defect of glass insulators is provided. Figure 1 The server in the example is used to illustrate the following steps:
[0111] Step S901, obtaining abnormal images corresponding to glass insulators of different orders, and obtaining normal images corresponding to glass insulators of different orders.
[0112] Step S902, performing feature extraction processing on the abnormal image to obtain the steel cap group features of the abnormal image corresponding to the first glass insulator, the abnormal gap features of the abnormal image corresponding to the middle glass insulator, and the connection fittings and tail fittings features of the abnormal image corresponding to the last glass insulator.
[0113] Step S903, determining the abnormal category of the glass insulator according to the steel cap group characteristics, abnormal gap characteristics, and connection hardware and tail hardware characteristics.
[0114] Step S904: annotate the abnormal images according to the abnormal categories to obtain abnormal image samples corresponding to the abnormal categories, and annotate the normal images as normal categories to obtain normal image samples corresponding to the normal categories.
[0115] Step S905, obtaining an original self-destruction defect recognition model according to the convolutional network backbone and the feature fusion unit; fusing the multi-head self-attention network and the original self-destruction defect recognition model to obtain an initial self-destruction defect recognition model.
[0116] Step S906: input the abnormal image samples and the normal image samples into the initial self-explosion defect recognition model to obtain the predicted categories corresponding to the abnormal image samples and the predicted categories corresponding to the normal image samples.
[0117] Step S907, iteratively training the initial self-destruction defect recognition model according to the category difference between the predicted category corresponding to the abnormal image sample and the abnormal category corresponding to the abnormal image sample, and the category difference between the predicted category corresponding to the normal image sample and the normal category, to obtain a trained self-destruction defect recognition model.
[0118] Step S908, obtaining inspection images of a plurality of continuous glass insulators; performing self-explosion defect recognition processing on the inspection images through the trained self-explosion defect recognition model to obtain the self-explosion defect recognition result of each glass insulator.
[0119] The above-mentioned method for identifying self-explosion defects of glass insulators can achieve the following beneficial effects: by enhancing the global information processing capability of the initial self-explosion defect recognition model, the self-explosion defect recognition capability of the enhanced self-explosion defect recognition model is effectively improved; by collecting inspection images of multiple consecutive glass insulators for self-explosion defect recognition processing, the self-explosion defects of glass insulators of different orders can be more accurately identified, thereby further improving the accuracy of self-explosion defect recognition of glass insulators.
[0120] In order to more clearly illustrate the method for identifying the self-explosion defect of a glass insulator provided by the embodiment of the present disclosure, the method for identifying the self-explosion defect of a glass insulator is specifically described below with a specific embodiment. Fig.10 As shown in the figure, another method for identifying the self-explosion defect of glass insulators is provided, which can be applied to Figure 1 The server in the system includes the following:
[0121] The multi-angle and multi-scale historical inspection images of glass insulators on the transmission line are taken by drones. The historical inspection images are labeled by categories to obtain labeled image samples; the labeled categories include the top steel cap of the glass insulator, the abnormal gap of the glass insulator, the self-explosion steel cap of the glass insulator, the self-explosion gap at the bottom of the glass insulator and normal. Based on the convolutional neural network and the multi-head attention network, the initial self-explosion defect recognition model is constructed to enhance the detection of the self-explosion of the first and last glass insulators. The initial self-explosion defect recognition model is trained with image samples to obtain the trained self-explosion defect recognition model. The current inspection image of the glass insulator is taken by drones, and then the current inspection image is input into the trained self-explosion defect recognition model to obtain the self-explosion defect recognition result of the current inspection image. The self-explosion defect recognition result can also be logically optimized to output the target recognition result.
[0122] In this embodiment, by enhancing the global information processing capability of the initial self-explosion defect recognition model, the self-explosion defect recognition capability of the enhanced self-explosion defect recognition model is effectively improved. By collecting inspection images of multiple consecutive glass insulators for self-explosion defect recognition processing, the self-explosion defects of glass insulators in different orders can be more accurately identified, and the detection of the self-explosion of the first glass insulator and the last glass insulator is enhanced, thereby reducing false detection and missed detection of glass insulator self-explosion, thereby further improving the accuracy of self-explosion defect recognition of glass insulators.
[0123] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0124] Based on the same inventive concept, the embodiment of the present application also provides a glass insulator self-explosion defect identification device for implementing the above-mentioned glass insulator self-explosion defect identification method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more embodiments of the glass insulator self-explosion defect identification device provided below can refer to the limitations of the glass insulator self-explosion defect identification method above, and will not be repeated here.
[0125] In one embodiment, Fig.11 As shown, a glass insulator self-explosion defect recognition device 1100 is provided, comprising: a model acquisition module 1101, an image acquisition module 1102 and a defect recognition module 1103, wherein:
[0126] The model acquisition module 1101 is used to acquire a trained self-detonation defect recognition model; the trained self-detonation defect recognition model is obtained by training an initial self-detonation defect recognition model through abnormal image samples corresponding to abnormal categories and normal image samples corresponding to normal categories; the abnormal categories are obtained according to abnormal feature settings corresponding to glass insulators in different orders; the initial self-detonation defect recognition model is obtained by performing global information enhancement processing on the original self-detonation defect recognition model.
[0127] The image acquisition module 1102 is used to acquire inspection images of a plurality of continuous glass insulators.
[0128] The defect recognition module 1103 is used to perform self-destruction defect recognition processing on the inspection image through the trained self-destruction defect recognition model to obtain the self-destruction defect recognition result of each glass insulator.
[0129] In one embodiment, the model acquisition module 1101 is also used to obtain abnormal image samples corresponding to abnormal categories of continuous glass insulators and normal image samples corresponding to normal categories of continuous glass insulators; the abnormal image samples and the normal image samples are input into the initial self-detonation defect recognition model to obtain the predicted categories corresponding to the abnormal image samples and the predicted categories corresponding to the normal image samples; according to the category difference between the predicted categories corresponding to the abnormal image samples and the abnormal categories corresponding to the abnormal image samples, and the category difference between the predicted categories corresponding to the normal image samples and the normal categories, the initial self-detonation defect recognition model is iteratively trained to obtain a trained self-detonation defect recognition model.
[0130] In one embodiment, the glass insulator self-explosion defect identification device 1100 also includes a sample annotation module, which is used to obtain abnormal images corresponding to glass insulators of different orders, and to obtain normal images corresponding to glass insulators of different orders; perform feature extraction processing on the abnormal images to obtain steel cap group features of the abnormal image corresponding to the first glass insulator, abnormal gap features of the abnormal image corresponding to the middle glass insulator, and connection fittings and tail fittings features of the abnormal image corresponding to the last glass insulator; determine the abnormal category of the glass insulator according to the steel cap group features, abnormal gap features, and connection fittings and tail fittings features; annotate the abnormal images according to the abnormal categories to obtain abnormal image samples corresponding to the abnormal categories, and annotate the normal images as normal categories to obtain normal image samples corresponding to the normal categories.
[0131] In one embodiment, the glass insulator self-explosion defect identification device 1100 also includes a model improvement module, which is used to obtain an original self-explosion defect identification model based on a convolutional network backbone and a feature fusion unit; and fuse the multi-head self-attention network and the original self-explosion defect identification model to obtain an initial self-explosion defect identification model.
[0132] In one embodiment, the defect recognition module 1103 is also used to perform feature extraction processing and global response normalization processing based on self-supervised learning on the inspection image through the convolutional network backbone in the self-destruction defect recognition model to obtain feature data of the inspection image; perform global information analysis processing on the feature data through the multi-head self-attention network in the self-destruction defect recognition model to obtain updated feature data of the inspection image; perform feature fusion processing on the updated feature data through the feature fusion unit in the self-destruction defect recognition model to obtain category recognition results of the inspection image; and obtain the self-destruction defect recognition results of each glass insulator according to the category recognition results of the inspection image corresponding to each glass insulator.
[0133] In one embodiment, the glass insulator self-explosion defect identification device 1100 also includes a feature extraction module, which is used to perform mask processing on the inspection image through a convolutional network backbone to obtain a mask image of the inspection image; perform mask prediction processing on the mask image through the encoder and decoder of the convolutional network backbone to obtain a predicted image corresponding to the mask image; the predicted image is used to characterize the predicted data of the mask area image in the mask image; and the predicted image, the mask image and the inspection image are sequentially subjected to global feature aggregation processing, feature normalization processing and feature calibration processing to obtain feature data.
[0134] Each module in the above-mentioned glass insulator self-explosion defect identification device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.
[0135] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.12As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as abnormal image samples, normal image samples and inspection images. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for identifying self-explosion defects of glass insulators is implemented.
[0136] Those skilled in the art will understand that Fig.12 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0137] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.
[0138] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.
[0139] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.
[0140] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.
[0141] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0142] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A method for identifying a self-explosion defect of a glass insulator, characterized in that: The method comprises: Acquire a trained self-detonation defect recognition model; the trained self-detonation defect recognition model is obtained by training an initial self-detonation defect recognition model through abnormal image samples corresponding to abnormal categories and normal image samples corresponding to normal categories; the abnormal categories are obtained according to abnormal feature settings corresponding to glass insulators of different orders; the initial self-detonation defect recognition model is obtained by performing global information enhancement processing on the original self-detonation defect recognition model; Obtain inspection images of multiple consecutive glass insulators; The self-explosion defect recognition model that has been trained is used to perform self-explosion defect recognition processing on the inspection image to obtain a self-explosion defect recognition result for each glass insulator.
2. The method according to claim 1, characterized in that The step of obtaining the trained self-destruction defect recognition model includes: Acquire abnormal image samples of continuous glass insulators corresponding to the abnormal category and normal image samples of continuous glass insulators corresponding to the normal category; Inputting the abnormal image sample and the normal image sample into the initial self-explosion defect recognition model to obtain the predicted category corresponding to the abnormal image sample and the predicted category corresponding to the normal image sample; According to the category difference between the predicted category corresponding to the abnormal image sample and the abnormal category corresponding to the abnormal image sample, and the category difference between the predicted category corresponding to the normal image sample and the normal category, the initial self-destruction defect recognition model is iteratively trained to obtain the trained self-destruction defect recognition model.
3. The method according to claim 2, characterized in that The step of acquiring abnormal image samples corresponding to the abnormal category of continuous glass insulators and normal image samples corresponding to the normal category of continuous glass insulators includes: Acquiring abnormal images corresponding to the glass insulators of different orders, and acquiring normal images corresponding to the glass insulators of different orders; Performing feature extraction processing on the abnormal image to obtain steel cap group features of the abnormal image corresponding to the first glass insulator, abnormal gap features of the abnormal image corresponding to the middle glass insulator, and connection fittings and tail fittings features of the abnormal image corresponding to the last glass insulator; Determine the abnormality category of the glass insulator according to the characteristics of the steel cap group, the characteristics of the abnormal gap, and the characteristics of the connecting hardware and the tail hardware; According to the abnormal category, the abnormal image is labeled to obtain abnormal image samples corresponding to the abnormal category, and the normal image is labeled as a normal category to obtain normal image samples corresponding to the normal category.
4. The method according to claim 2, characterized in that: Before inputting the abnormal image sample and the normal image sample into the initial self-explosion defect recognition model to obtain the predicted category corresponding to the abnormal image sample and the predicted category corresponding to the normal image sample, the method further includes: According to the convolutional network backbone and the feature fusion unit, the original self-explosion defect recognition model is obtained; The multi-head self-attention network and the original self-destruction defect recognition model are fused to obtain the initial self-destruction defect recognition model.
5. The method according to any one of claims 1 to 4, characterized in that: The self-explosion defect recognition model completed through training performs self-explosion defect recognition processing on the inspection image to obtain the self-explosion defect recognition result of each glass insulator, including: Through the convolutional network backbone in the self-explosion defect recognition model, the inspection image is subjected to feature extraction processing and global response normalization processing based on self-supervised learning to obtain feature data of the inspection image; Performing global information analysis and processing on the feature data through the multi-head self-attention network in the self-explosion defect recognition model to obtain updated feature data of the inspection image; The updated feature data is subjected to feature fusion processing by a feature fusion unit in the self-explosion defect recognition model to obtain a category recognition result of the inspection image; According to the category recognition result of the inspection image corresponding to each glass insulator, the self-explosion defect recognition result of each glass insulator is obtained.
6. The method according to claim 5, characterized in that The method further comprises: performing feature extraction processing and global response normalization processing based on self-supervised learning on the inspection image through the convolutional network backbone in the self-explosion defect recognition model to obtain feature data of the inspection image. Through the convolutional network backbone, mask processing is performed on the inspection image to obtain a mask image of the inspection image; Through the encoder and decoder of the convolutional network backbone, the mask image is subjected to mask prediction processing to obtain a predicted image corresponding to the mask image; the predicted image is used to represent the predicted data of the mask area image in the mask image; The predicted image, the mask image and the inspection image are sequentially subjected to global feature aggregation processing, feature normalization processing and feature calibration processing to obtain the feature data.
7. A glass insulator self-explosion defect identification device, characterized in that: The device comprises: The model acquisition module is used to acquire a trained self-detonation defect recognition model; the trained self-detonation defect recognition model is obtained by training an initial self-detonation defect recognition model through abnormal image samples corresponding to abnormal categories and normal image samples corresponding to normal categories; the abnormal categories are obtained according to abnormal feature settings corresponding to glass insulators of different orders; the initial self-detonation defect recognition model is obtained by performing global information enhancement processing on the original self-detonation defect recognition model; An image acquisition module is used to acquire inspection images of multiple consecutive glass insulators; The defect recognition module is used to perform self-explosion defect recognition processing on the inspection image through the trained self-explosion defect recognition model to obtain the self-explosion defect recognition result of each glass insulator.
8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.