Insulator defect detection method, device and equipment and storage medium

Through the combination of feature extraction enhancement module, insulator segmentation task head network and fusion network, combined with insulator detection task head network, the problems of low real-time and accuracy in insulator defect detection methods are solved, and efficient detection in complex backgrounds is achieved.

CN119067930APending Publication Date: 2024-12-03GUANGDONG POWER GRID CO LTD +1

Patent Information

Application Number
CN202411108726.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-13
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing insulator defect detection methods have problems with low real-time and low accuracy, especially in complex background environments, which are difficult to achieve real-time and high-accuracy detection.

Method used

Feature extraction enhancement module is used to extract and enhance the image to be detected, and multi-scale feature maps are segmented and fused through insulator segmentation task head network and fusion network, and defect detection is performed in combination with insulator detection task head network to improve the real-time and accuracy of detection. Rate.

Benefits of technology

The feature map is segmented and detected by different task heads, which significantly improves the real-time and accuracy of insulator defect detection, solves the problem of low real-time and accuracy in the prior art, and is suitable for detection applications in complex background environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119067930A_ABST
    Figure CN119067930A_ABST
Patent Text Reader

Abstract

The invention discloses an insulator defect detection method, device and equipment and a storage medium. The method comprises the following steps: performing feature extraction and enhancement on a to-be-detected image through a feature extraction and enhancement module to obtain a multi-scale feature map corresponding to the to-be-detected image; segmenting the to-be-detected image based on shallow layer features in the multi-scale feature map through an insulator segmentation task head network to obtain an intermediate feature map; fusing the feature map of each scale in the multi-scale feature map with the intermediate feature map through a fusion network to obtain fused feature maps; and detecting the fused feature maps through an insulator detection task head network to obtain a defect detection result of the to-be-detected image. According to the method, an end-to-end network structure with multiple task heads is constructed, and segmentation and feature detection are simultaneously carried out on the content of a to-be-detected image based on a feature map; and the task head network trained by adopting a multi-task learning method is used, so that the real-time performance and accuracy of defect detection are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of defect detection technology, and in particular to an insulator defect detection method, device, equipment and storage medium. Background Art

[0002] Insulators are important components of the power transmission and distribution network system. They are responsible for supporting and fixing transmission lines and preventing current from returning to the ground. Since insulators are exposed to the outdoor environment for a long time, and are subjected to harsh natural conditions such as strong winds, heavy rains, and blizzards, as well as the working environment of high-voltage power transmission and lightning strikes, insulators are prone to defects such as flashing and breakage, which can lead to power grid tripping accidents, affect the normal operation of the power transmission network, and bring serious economic losses and inconvenience to life. As an important insulator inspection method, drone inspection has been widely used. Combined with intelligent image analysis algorithms, it can greatly improve the detection efficiency of insulator defects. Existing insulator defect detection algorithms can be divided into two categories: end-to-end methods and two-stage methods.

[0003] The two-stage insulator detection algorithm first extracts each insulator string area from the collected transmission and distribution network inspection image, then determines whether the extracted insulator string has defects, and classifies and locates the defects. In the first stage, the insulator string area is extracted, and the background area is reduced to reduce the interference of the background on the next defect detection, which has a high accuracy. However, the two stages of the two-stage insulator detection algorithm are independent neural networks and algorithms, which are highly complex and consume a lot of computing resources on the end-side equipment, resulting in high inference delays and difficulty in real-time detection of insulator defects. End-to-end insulator defect detection algorithms usually use target detection technology to detect the location and category of insulator defects at the same time. Because it uses an end-to-end neural network, it can effectively fuse image information and has the characteristics of fast inference speed, which is suitable for end-side insulator detection applications with high real-time requirements. However, the end-to-end insulator defect detection algorithm is often affected by the complex background environment during application, and the defect detection accuracy is not as good as the two-stage method. Summary of the invention

[0004] The present invention provides an insulator defect detection method, device, equipment and storage medium to solve the problems of low real-time performance and low accuracy of the insulator defect detection method in the prior art.

[0005] According to one aspect of the present invention, a method for detecting insulator defects is provided, the method comprising:

[0006] The feature extraction and enhancement module is used to extract and enhance the features of the image to be detected, thereby obtaining a multi-scale feature map corresponding to the image to be detected;

[0007] Segmenting the image to be detected based on shallow features in the multi-scale feature map through an insulator segmentation task head network to obtain an intermediate feature map;

[0008] The feature map of each scale in the multi-scale feature map is respectively fused with the intermediate feature map through a fusion network to obtain each fused feature map;

[0009] The fused feature maps are detected through the insulator detection task head network to obtain the defect detection result of the image to be detected.

[0010] According to another aspect of the present invention, there is provided an insulator defect detection device, the device comprising:

[0011] A feature extraction and enhancement module is used to extract and enhance features of the image to be detected to obtain a multi-scale feature map corresponding to the image to be detected;

[0012] An insulator segmentation task head network is used to segment the image to be detected based on shallow features in the multi-scale feature map to obtain an intermediate feature map;

[0013] A fusion network, used for fusing the feature map of each scale in the multi-scale feature map with the intermediate feature map to obtain each fused feature map;

[0014] The insulator detection task head network is used to detect the fused feature maps to obtain the defect detection result of the image to be detected.

[0015] According to another aspect of the present invention, there is provided an electronic device, the electronic device comprising: at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the insulator defect detection method described in any embodiment of the present invention.

[0018] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the insulator defect detection method described in any embodiment of the present invention when executed.

[0019] An insulator defect detection method, device, equipment and storage medium according to an embodiment of the present invention, the method includes: extracting and enhancing features of an image to be detected by a feature extraction and enhancement module to obtain a multi-scale feature map corresponding to the image to be detected; segmenting the image to be detected based on shallow features in the multi-scale feature map by an insulator segmentation task head network to obtain an intermediate feature map; fusing the feature map of each scale in the multi-scale feature map with the intermediate feature map by a fusion network to obtain each fused feature map; and detecting each fused feature map by an insulator detection task head network to obtain a defect detection result of the image to be detected. The method can improve the real-time and accuracy of defect detection by segmenting and detecting feature maps with different task heads, and solves the problem of low real-time and accuracy of insulator defect detection methods in the prior art.

[0020] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 creative work.

[0022] Figure 1 A schematic diagram of a flow chart of an insulator defect detection method provided in Embodiment 1 of the present invention;

[0023] Figure 2 A schematic diagram of a multi-scale feature map provided by an embodiment of the present invention;

[0024] Figure 3 A network architecture diagram of an insulator defect detection method provided by an embodiment of the present invention;

[0025] Figure 4 A schematic diagram of the architecture of a task head module provided by an embodiment of the present invention;

[0026] Figure 5 A schematic diagram of the structure of an insulator segmentation task head network provided by an embodiment of the present invention;

[0027] Figure 6 A schematic diagram of a CBS structure provided by an embodiment of the present invention;

[0028] Figure 7A schematic flow chart of an insulator defect detection method provided in Embodiment 2 of the present invention;

[0029] Figure 8 A schematic diagram of a process for fusing an intermediate feature map and a feature map at each scale provided in an embodiment of the present invention;

[0030] Fig. 9 A schematic diagram of the structure of an insulator detection task head network provided by an embodiment of the present invention;

[0031] Fig.10 A schematic structural diagram of an insulator defect detection device provided in Embodiment 3 of the present invention;

[0032] Fig.11 Schematic diagram of the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only an embodiment of a part of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should belong to the scope of protection of the present invention. It should be understood that the various steps recorded in the method implementation of the present invention can be performed in different orders and / or in parallel. In addition, the method implementation may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0034] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0035] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, any variations of the terms "including" and "having" are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0036] It should be noted that the modifications of "one" and "plurality" mentioned in the present invention are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0037] The names of the messages or information exchanged between multiple devices in the embodiments of the present invention are only used for illustrative purposes, and are not used to limit the scope of these messages or information.

[0038] Embodiment 1

[0039] Figure 1 The present invention provides a flowchart of an insulator defect detection method provided in the first embodiment of the present invention. The method is applicable to the case of detecting defects in insulators. The method can be performed by an insulator defect detection device, wherein the device can be implemented by software and / or hardware and is generally integrated on an electronic device. In this embodiment, the electronic device includes but is not limited to: a computer and other devices. Figure 1 As shown, a method for detecting insulator defects provided by Embodiment 1 of the present invention comprises the following steps:

[0040] S110 , extracting and enhancing features of the image to be detected by a feature extraction and enhancement module to obtain a multi-scale feature map corresponding to the image to be detected.

[0041] The feature extraction and enhancement module may be a module for preprocessing the image to be detected, and the preprocessing may include feature extraction and enhancement. The image to be detected may be an image related to an insulator. A multi-scale feature map may be a technology used in computer vision and image processing tasks, and a multi-scale feature map may be used to capture information at different scales in an image. A multi-scale feature map is usually formed by superimposing feature maps of multiple scales, and each feature map of each scale corresponds to information at different levels or scales of the image.

[0042] In this embodiment, the feature extraction and enhancement module can be used to extract and enhance features of the image to be detected to obtain feature maps of the image to be detected at multiple scales. This embodiment does not limit the method for extracting and enhancing features of the image to be detected. Specifically, it can be to first integrate the deep semantic information in the image into the shallow features in a top-down manner, and then further propagate the main features such as shallow image texture to the deep layer of the network based on a bottom-up manner for feature fusion, and finally form an effective fused multi-scale feature map.

[0043] The multi-scale feature map in this embodiment may include three scales, which can be divided into shallow features, middle features and deep features according to the size of each layer from large to small. Shallow features generally refer to the features extracted by the lower layers of the network (close to the input layer). These features generally include basic image features such as edges, corners, colors, textures, etc. Middle-level features are features extracted by the middle layers (hidden layers) of the network, which are usually located between shallow features and deep features. Middle-level features are a combination of shallow features and are often used to capture more complex shapes, patterns and local structures. Deep features are features extracted by the higher layers of the network (close to the output layer), which generally contain high-level concepts or semantic information. These features may represent specific object categories, attributes or more abstract concepts. Exemplarily, Figure 2 A schematic diagram of a multi-scale feature map provided by an embodiment of the present invention is shown in FIG. Figure 2 As shown, the feature includes three scales, which are the feature map results of 8 times downsampling, 16 times downsampling and 32 times downsampling of the image to be detected.

[0044] S120, segmenting the image to be detected based on shallow features in the multi-scale feature map through the insulator segmentation task head network to obtain an intermediate feature map.

[0045] Among them, the insulator segmentation task head network can be used to segment the input image to be detected based on the feature map. The task head refers to a part in the deep learning model that is responsible for the output layer or subnetwork of a specific task. The task head network is a module composed of one or more task heads, usually in conjunction with a shared basic feature network. The intermediate feature map can be the output result of the intermediate layer of the insulator segmentation task head network when processing shallow features.

[0046] In this embodiment, the shallow feature map has a large size and contains more information such as image texture, and is more suitable for segmentation tasks. Therefore, the shallow features can be input into the insulator segmentation task head network. The insulator segmentation task head network can segment the image to be detected based on the shallow features in the multi-scale feature map to obtain an intermediate feature map.

[0047] S130, fusing the feature map of each scale in the multi-scale feature map with the intermediate feature map through a fusion network to obtain fused feature maps.

[0048] Among them, the fusion network can be used to fuse different feature maps.

[0049] In this embodiment, the fusion network can receive the intermediate feature map output by the insulator segmentation task head network, as well as the feature maps at different scales output by the feature extraction enhancement module. For the feature maps at different scales, the intermediate feature map can be fused with the feature maps at different scales through the fusion network to obtain multiple fused feature maps.

[0050] S140. Detect each fused feature map through the insulator detection task head network to obtain a defect detection result of the image to be detected.

[0051] Among them, the insulator detection task head network can be used to detect defects of insulators to obtain detection results. The defect detection results can include information such as the positioning box information, category, and confidence level of the included object of the insulator in the image.

[0052] In this embodiment, each fused feature map can be detected by the insulator detection task head network to obtain detection results at different scales. The detection results at different scales can be comprehensively analyzed to obtain the final defect detection result of the image to be detected. The method of analyzing the detection results at different scales is not described in detail in this embodiment. The identification and positioning of insulator defect types requires not only image texture information, but also deeper semantic information. Therefore, in this embodiment, the features at three scales can be input into the insulator detection task head network.

[0053] For example, Figure 3A network architecture diagram of an insulator defect detection method provided in an embodiment of the present invention. The insulator defect detection method in this embodiment can be implemented by two modules: a feature extraction enhancement module and a task head network. The feature extraction enhancement module can effectively analyze the input image and obtain image features at multiple scales. The downstream task head module can detect and analyze the insulator, and the insulator segmentation task head network and the insulator detection task head network are two related tasks. The task head module also designs a fusion network, which uses the effective features of the middle layer of the insulator segmentation task head network to eliminate the interference of the background on the insulator detection task head network recognition. The multi-task head network structure of the present invention can support the simultaneous segmentation and detection tasks, taking into account the real-time and accuracy of end-to-end reasoning, and avoiding the inference inefficiency caused by data handling during the stage conversion process of the multi-stage method. The feature extraction enhancement module of the present invention can adopt the backbone network and neck structure of YOLOv7 (YouOnly Look Once version 7) to obtain multi-scale features of information fusion.

[0054] In one embodiment, the above architecture can be deployed on the end-side device. The end-side device refers to the device or node close to the data source and the user, emphasizing that part of the data processing and storage functions are placed on these devices to improve the response speed and reduce the network load.

[0055] In this embodiment, by applying the above method to the end-side device, the data transmission process can be reduced, the inference delay can be reduced, and the detection accuracy can be improved. For example, it can be deployed by building an embedded system on the end-side device, such as carrying an embedded computing system on a drone, and the computing system can be used to directly process (image) data on the drone.

[0056] Figure 4 A schematic diagram of the architecture of a task header module provided by an embodiment of the present invention, such as Figure 4 As shown, the task head module includes an insulator segmentation task head network, an insulator detection task head network and a fusion network. The fusion network can fuse feature maps at different scales with the intermediate feature maps output by the insulator segmentation task head network, and then send the fused feature maps to the insulator detection task head network.

[0057] An insulator defect detection method provided in the first embodiment of the present invention includes: extracting and enhancing features of an image to be detected by a feature extraction and enhancement module to obtain a multi-scale feature map corresponding to the image to be detected; segmenting the image to be detected based on shallow features in the multi-scale feature map by an insulator segmentation task head network to obtain an intermediate feature map; fusing the feature map of each scale in the multi-scale feature map with the intermediate feature map by a fusion network to obtain each fused feature map; and detecting each fused feature map by an insulator detection task head network to obtain a defect detection result of the image to be detected. The method can improve the real-time and accuracy of defect detection by segmenting and detecting feature maps with different task heads, and solves the problem of low real-time and accuracy of insulator defect detection methods in the prior art.

[0058] The end-to-end neural network method of this embodiment adopts the neural network structure paradigm of "feature extraction and enhancement" and "task head", designs a new "task head" module network structure, and proposes an insulator and defect type detection network based on multi-task learning. The "head" of the "task head" network has two branches, which respectively perform the insulator segmentation task and the insulator defect type detection task, and has the ability of multi-task joint learning, which can comprehensively improve the detection accuracy.

[0059] Based on the above embodiment, a variant embodiment of the above embodiment is proposed. It should be noted that in order to make the description concise, only the differences from the above embodiment are described in the variant embodiment.

[0060] In one embodiment, the insulator segmentation task head network segments the image to be detected based on shallow features in the multi-scale feature map to obtain an intermediate feature map, including: obtaining shallow features in the multi-scale feature map; performing a convolution operation on the shallow features through the insulator segmentation task head network to obtain an intermediate feature map corresponding to the image to be detected.

[0061] Among them, in convolutional neural networks, convolution operation is an important operation used to extract features from input images.

[0062] In this embodiment, after obtaining shallow features in the multi-scale feature map through the insulator segmentation task head network, one or more convolution operations can be performed on the shallow features to obtain an intermediate feature map in the process of segmenting the image to be detected.

[0063] For example, Figure 5 A schematic diagram of the structure of an insulator segmentation task head network provided by an embodiment of the present invention, such as Figure 5As shown in the figure, the insulator segmentation task head network takes the enhanced shallow features as input and the CBS structure as the basic component. The enhanced shallow features input by the network can first be transformed into features through four CBS structures. Figure 1 , and then through the second set of 4 CBS structural features Figure 2 Finally, after a CBS structure, 2 times upsampling and a 1x1 convolution layer, a one-dimensional feature segmentation mask is finally formed. The segmentation mask can express the segmentation result of the insulator area. Figure 2 As an intermediate feature map, the segmentation mask can be used when training the head network for the insulator segmentation task. Figure 3 The output segmentation mask is twice the size of the input shallow feature and 1 / 4 the size of the input image, which can form a larger segmentation mask and provide more accurate segmentation results. Limited upsampling (two times upsampling) can help improve the accuracy of the segmentation results and bring better segmentation features. On the other hand, the smaller segmentation mask will be used with the true value of the segmentation to construct the loss function for model training, which will help reduce the overall computational requirements of the network.

[0064] The CBS module is the most basic network building module in the YOLOv7 network, which consists of Convolution+Batch-Normalization+SiLU. CBS has been widely used in YOLO v7 and can significantly improve the accuracy of target detection tasks. Figure 6 A schematic diagram of a CBS structure provided by an embodiment of the present invention is shown in FIG. Figure 6 As shown, the CBS structure includes a convolution layer, a batch normalization layer, and a SiLu activation layer. The size and step of the convolution kernel can be customized. The optional convolution kernel size includes but is not limited to 1x1, 3x3, 5x5, and the optional step includes but is not limited to 1 and 2.

[0065] In one embodiment, the feature extraction and enhancement module also includes a feature extraction network and a feature enhancement module. The image to be detected first passes through the feature extraction network to form three different scale features of the first level; the sizes of these three scale features are 1 / 8, 1 / 16, and 1 / 32 of the input image size, respectively.

[0066] The feature extraction network includes 4 CBS structures, 3 groups of ELAN (Effective Long-Range Aggregation Network) and MP-1 combined structures, and 1 ELAN structure. The CBS layer is the basic structure of the feature extraction network. It is a series structure consisting of convolution, BN layer, and SiLU activation layer. The convolution kernel size and step size can be set as needed. The ELAN structure is a composite structure composed of CBS as a component and is the core structure of feature extraction. MP-1 and MP-2 have the same structure, both consisting of CBS layers and maxpool operations. The difference is that the number of output channels of MP-1 is the same as the input, and the number of output channels of MP-2 is twice the input.

[0067] The three different scale features of the first level will enter the feature enhancement module. The feature enhancement module is composed of a Feature Pyramid Network (FPN) structure and a Path Aggregation Network (PAN) structure in series. FPN adopts a top-down feature fusion structure. First, the small-scale deep features of the first level are fused with the medium-scale middle-level features of the first level through the SPPCSPC (Spatial Pyramid Pooling, Cross Stage Partial Channel) structure and the CBS structure, and then upsampled by 2 times; the fused features are sequentially spliced ​​with the large-scale shallow features of the first level through the ELAN-W, CBS structure and 2-part upsampling, and then fused through ELAN-W. FPN integrates the semantic information of the top-level features into the low-level feature layers such as texture and color.

[0068] The features generated by FPN will be further fed into the PAN network, propagating low-level image features from bottom to top to high-level feature layers with rich semantic information. Specifically, the large-scale latent features generated by FPN are used as enhanced shallow-layer feature outputs for downstream tasks, and on the other hand, through the MP-2 structure, features of 1 / 2 times the input size of the MP-2 module are formed, and fused with the medium-sized middle-level features generated by FPN to form enhanced middle-level features; the fused results are passed through the ELAN-W and MP-2 structures, and then spliced ​​with the small-scale deep features generated by FPN, and finally fused through ELAN-W to form enhanced deep features.

[0069] In one embodiment, the enhanced multi-scale features pass through the task head module to form prediction results of three feature layers respectively. The detection results generated by each scale feature layer can include 255 channels. It can be divided into three groups, each with 85 channels, and the data size of the corresponding detection results is expressed as (N, 20, 20, 3, 85), (N, 40.40, 3, 85), (N, 80, 80, 3, 85), where N is the number of input images. Among the 85 channels, the first 4 channels correspond to the parameters of the prediction box; the 5th channel corresponds to the confidence that the prediction box contains an object; the last 80 channels are used to determine the category of the object contained in the target box. In an embodiment, the last 80 channel parameters correspond to 80 target categories, which are for the COCO (Common Objects in Context) dataset.

[0070] The position and size of the prediction box calculated in the above way are normalized values ​​relative to the width and height of the image. Therefore, the target boxes obtained by each feature layer can be summarized as candidate prediction results. All candidate target boxes can be further filtered by scores and non-maximum suppression to obtain the final prediction result.

[0071] Embodiment 2

[0072] Figure 7 This is a flow chart of an insulator defect detection method provided by the second embodiment of the present invention. The second embodiment is optimized on the basis of the above embodiments. For details not yet detailed in this embodiment, please refer to the first embodiment. Figure 7 As shown, a method for detecting insulator defects provided by Embodiment 2 of the present invention comprises the following steps:

[0073] S210 , extracting and enhancing features of the image to be detected by a feature extraction and enhancement module to obtain a multi-scale feature map corresponding to the image to be detected.

[0074] S220, segmenting the image to be detected based on shallow features in the multi-scale feature map through the insulator segmentation task head network to obtain an intermediate feature map.

[0075] S230. For each scale feature map in the multi-scale feature map, adjust the scale of the intermediate feature map to a scale matching the feature map by a feature scale adaptation module.

[0076] The feature scale adaptation module may be a module for converting the scale of a feature map.

[0077] In this embodiment, the scale of the intermediate feature map can also be adjusted to a scale matching the scale of the feature map of each scale through the feature scale adaptation module to facilitate subsequent processing.

[0078] S240. Fusing the adjusted intermediate feature map with the feature map through a fusion network to obtain a fused feature map.

[0079] In this embodiment, the multiple intermediate feature maps after scale adjustment can be fused with the feature maps at the corresponding scales through a fusion network to obtain multiple fused feature maps.

[0080] For example, Figure 8 A schematic diagram of a process for fusing an intermediate feature map and a feature map at each scale provided by an embodiment of the present invention, wherein the fusion network includes two main input branches, which respectively receive the intermediate features of the insulator segmentation task head network and the enhanced features output by the feature extraction enhancement module. Figure 8 As shown in the figure, the above two types of features are subjected to a convolution operation in each branch to obtain feature maps of the same size and dimension; then, after a splicing operation and a convolution operation along the feature dimension, they are fused in the feature dimension and space to obtain the fusion result of the feature maps of the two branches, which are input into the insulator detection task head network to carry out the detection of insulator defects. The spliced ​​features are then subjected to a convolution operation, and the characteristics of convolution operating on the feature channel dimension can be used to further fuse the information between channels to form a fusion feature.

[0081] This embodiment selects the features Figure 2 As an intermediate feature map, because in the insulator segmentation task head network, relative to the features obtained by upsampling (interpolation technique) Figure 3 , its characteristics Figure 2 The original information is kept, avoiding the error caused by upsampling (interpolation technology). At the same time, the segmentation mask results predicted by the insulator segmentation task head network often have mis-segmentation. Features Figure 2 As an intermediate feature, the extracted features are more universal and can tolerate errors in insulator region segmentation to a certain extent, which helps to accurately eliminate the interference of the insulator background and obtain accurate insulator identification and positioning results.

[0082] The fusion network includes a feature scale adaptation module. The feature scale adaptation module can adapt the size of the feature map from the insulator segmentation task head network to the size of the enhanced feature for splicing and fusion. Specifically, the size of the insulator segmentation network feature map can be adjusted by combining upsampling, downsampling and convolution calculation.

[0083] S250. Detect each fused feature map through the insulator detection task head network to obtain a defect detection result of the image to be detected.

[0084] The second embodiment of the present invention provides an insulator defect detection method, which specifically fuses the feature map of each scale in the multi-scale feature map with the intermediate feature map through a fusion network to obtain each fused feature map, including: for the feature map of each scale in the multi-scale feature map, adjusting the scale of the intermediate feature map to a scale matching the feature map through a feature scale adaptation module; fusing the adjusted intermediate feature map with the feature map through a fusion network to obtain a fused feature map. This embodiment further combines the intermediate feature map of the insulator segmentation task head network with the image enhancement features by constructing a fusion mechanism of the insulator segmentation task head network and the insulator detection task head network, thereby strengthening the insulator target area, eliminating background information irrelevant to the insulator, and reducing the interference of the background on the identification of the insulator defect type.

[0085] In one embodiment, the insulator detection task head network is used to detect each fused feature map to obtain a defect detection result of the image to be detected, including: for fused feature maps at different scales, the fused feature map is detected by the positioning branch and the classification branch in the insulator detection task head network respectively to obtain the positioning frame information of the fused feature map and the corresponding object confidence and category; the positioning frame in the positioning frame information that contains an object confidence that meets the requirements is used as the defect detection result of the image to be detected.

[0086] The positioning branch can be used to process positioning tasks, and the classification branch can be used to process classification tasks. The positioning frame information can be the coordinate information of the positioning frame in the image. The category can be the classification of the target in the positioning frame, and the object inclusion confidence can specify the confidence that a certain object (such as an insulator) is included in the positioning frame.

[0087] In this embodiment, different branches can be used to locate and classify the fused feature map, obtain the positioning frame information of the fused feature map and the corresponding confidence and category of the contained object, and then the positioning frame can be screened by the confidence of the contained object to obtain the positioning frame that meets the requirements and use it as the defect detection result of the image to be detected. The insulator detection task head network contains dedicated conversion layers for classification and positioning tasks. The decoupled structure of classification and positioning enables each branch to focus on its own task features, which can solve the feature misalignment problem caused by the shared features of the two types of tasks and improve the focus ability and detection accuracy of each task.

[0088] In one embodiment, the fused feature map is detected by the positioning branch and the classification branch in the insulator detection task head network to obtain the positioning frame information of the fused feature map and the corresponding confidence and category of the included object, including: extracting the positioning feature corresponding to the positioning branch and the classification feature corresponding to the classification branch from the fused feature map; detecting the fused feature map based on the positioning feature by the positioning branch to obtain the positioning frame information of the fused feature map and the corresponding confidence of the included object; detecting the fused feature map based on the classification feature by the classification branch to obtain the category corresponding to the positioning frame information.

[0089] Among them, the classification feature may be a feature related to the classification task, and the positioning feature may be a feature related to the positioning task.

[0090] In this embodiment, for the positioning branch and the classification branch, features related to the tasks of each branch, namely, the positioning features and the classification features, can be first extracted from the fused feature map, and then each branch can perform its task through its own extracted features.

[0091] For example, Fig. 9 A schematic diagram of the structure of an insulator detection task head network provided by an embodiment of the present invention is shown in FIG. Fig. 9 As shown, the insulator detection task head network adopts the idea of ​​decoupling to divide the defect detection task into two branches: defect classification and positioning. The positioning branch and the classification branch have similar structures. Before the data is input into different branches, two convolution modules are used to extract features suitable for their respective tasks, and then the output results of recognition and positioning are generated through convolution. The classification branch outputs a multidimensional classification map. Each dimension corresponds to a category, with a total of k categories, and category k can be set according to actual conditions. For example, if the insulator defect involves two types of insulator defect types, insulator flash grid and insulator breakage, the classification map has two dimensions, which represent insulator flash grid and insulator breakage respectively. The positioning branch outputs two types of information: positioning frame information and object confidence. The output positioning frame information can include four dimensions, which respectively represent the horizontal coordinate and vertical coordinate of the center of the positioning frame, and express the width and height information of the positioning frame with the horizontal offset and vertical offset relative to the center point, or it can also be the information of the four vertex coordinates of the positioning frame, which is not limited in this embodiment. The object confidence corresponds to the possibility of containing a defective target at this time, and is expressed in one dimension. The fused features used by the insulator detection task head network combine the intermediate feature map output by the insulator segmentation task head network and the multi-scale feature map output by the feature extraction enhancement module, which can suppress the interference of background on the detection of insulator defect types.

[0092] In one embodiment, after obtaining the positioning frame information, the method further includes: deleting redundant positioning frames in the positioning frame information by using a non-maximum suppression algorithm.

[0093] Among them, the Non-Maximum Suppression (NMS) algorithm is an algorithm for removing non-maximum values, and is often used in edge detection and object recognition in computer vision.

[0094] In this embodiment, since there is a lot of redundancy in the positioning frame results, a non-maximum suppression algorithm can also be used to filter out high-confidence positioning results and remove positioning frames with serious overlap.

[0095] In one embodiment, the method also includes: performing multi-task joint training on the insulator segmentation task head network and the insulator detection task head network through a loss function; the loss function includes a target detection task loss function and a segmentation task loss function, and the detection task loss function includes a classification loss function, a confidence loss function and a target box loss function.

[0096] Among them, the loss function can be a function used to measure the degree of inconsistency between the predicted value of the model and the true value. The target detection task loss function can be a loss function used to train the head network of the insulator detection task, the classification loss function can be a loss function used to train the accuracy of the category output, the confidence loss function can be a loss function used to train the accuracy of the confidence output of the included object, and the target box loss function can be a loss function used to train the accuracy of the positioning box output. The segmentation task loss function can be a loss function used to train the head network of the insulator segmentation task. Multi-task joint training can refer to the presence of multiple subtasks in the model, and these subtasks are trained together.

[0097] In this embodiment, a multi-task learning method can be used to train the insulator segmentation task head network and the insulator detection task head network at the same time, and the following loss function can be constructed:

[0098] L=αL class +βL obj +γL bbox +λL seg

[0099] Among them, L class , L obj , L bbox The target detection task belongs to the insulator detection task head network, which corresponds to the classification, object confidence, and target box outputs of the decoupling head of the insulator detection task head network. segThe segmentation task of the insulator segmentation task head network. Among them, α, β, γ, λ are the weight coefficients of each loss function, which are used to coordinate each task. The optional set of parameter values ​​includes but is not limited to α=0.3, β=0.7, γ=0.05, λ=0.1. The weight coefficients of the above loss functions can also be normalized so that the sum of the weight coefficients is 1.

[0100] Classification loss function L in target detection task class as follows:

[0101]

[0102] Where N is the number of prediction positioning frames, K is the number of insulator defect categories, and y i is the true value vector of the category of the i-th predicted positioning box, y i =[y i0 ,y i1 ,…,y ic ,…y iK ],y ic ∈{0,1};p i is the predicted category probability vector of the i-th predicted positioning box, p i =[p i0 ,p i1 ,…,p ic ,…p iK ],0≤p ic ≤1;σ(x)=1 / (1+e -x ) is the Sigmoid function, and e is the natural base.

[0103] Given N prediction boxes, the confidence loss function L in the target detection task is obj as follows:

[0104]

[0105] Among them, t i is the IoU between the i-th predicted positioning box and the corresponding true positioning box, IoU = (predicted positioning box ∩ true positioning box) / (predicted positioning box ∪ true positioning box); i The confidence prediction value of the object contained in the i-th predicted positioning box is calculated; σ(x) is the Sigmoid function.

[0106] Target box loss function L in target detection task bbox It is expressed as follows:

[0107]

[0108] Among them, b i is the i-th predicted positioning box, B iis the corresponding real positioning box, “∩” is the intersection operation, and “∪” is the union operation.

[0109] Segmentation task loss function L seg It is expressed as follows:

[0110]

[0111] Where P is the total number of pixels in the mask image; m i is the label of the i-th pixel. When the i-th pixel belongs to the insulator region, m i =1, otherwise 0; l i is the predicted probability that the i-th pixel belongs to an insulator, 0≤l i ≤1;σ(x) is the Sigmoid function.

[0112] In the process of training the above network structure, a data set for network training can be constructed. Collect real images, clean and filter the images, and remove redundant and incomplete image data. The segmentation and annotation results of the insulator are obtained by drawing polygons; the annotation of two types of defects, insulator flash and insulator damage, are obtained by drawing rectangular boxes. The data annotation format can be PASCAL VOC (The PASCAL Visual Object Classes) format. The annotated image set can be divided into training set, verification set and test set.

[0113] To enhance the generalization of the training model, geometric operations such as horizontal and vertical flipping, random cropping, rotation and shifting of images can be used to increase the diversity of samples; data such as white noise can be added to the input image to enhance the robustness of the network. The network for model training can use the pre-trained model for parameter initialization, which helps to transfer the ability of the pre-trained model to capture effective features to the target task data set, so as to improve performance indicators and obtain faster convergence speed.

[0114] During the training process, the image data that has been enhanced can be sent to the network in batches for iterative training. Each batch of data is forward propagated to obtain the total loss function value; based on the loss function value of the batch of data, the gradient value of each trained parameter weight is calculated by back propagation; the model parameters are updated layer by layer by the loss function. After a certain number of iterations, the performance of the trained model is evaluated using the validation set to select the model, avoid overfitting, and improve the generalization ability of the model. The performance of the trained model will be qualitatively and quantitatively verified on the test set. Continue to adjust the network parameters, supplement and correct the data, and train the model until the model meets the required indicators.

[0115] Embodiment 3

[0116] Fig.10This is a schematic diagram of the structure of an insulator defect detection device provided in the third embodiment of the present invention. The device can be applied to the situation of detecting defects of insulators, wherein the device can be implemented by software and / or hardware and is generally integrated in an electronic device. Fig.10 As shown, the device comprises:

[0117] A feature extraction and enhancement module 310 is used to extract and enhance features of the image to be detected to obtain a multi-scale feature map corresponding to the image to be detected;

[0118] Insulator segmentation task head network 320, used to segment the image to be detected based on the shallow features in the multi-scale feature map to obtain an intermediate feature map;

[0119] A fusion network 330, used for fusing the feature map of each scale in the multi-scale feature map with the intermediate feature map to obtain fused feature maps;

[0120] The insulator detection task head network 340 is used to detect the fused feature images to obtain the defect detection result of the image to be detected.

[0121] The present embodiment provides an insulator defect detection device, the device comprising: a feature extraction and enhancement module, used to extract and enhance features of an image to be detected, and obtain a multi-scale feature map corresponding to the image to be detected; an insulator segmentation task head network, used to segment the image to be detected based on shallow features in the multi-scale feature map, and obtain an intermediate feature map; a fusion network, used to fuse the feature map of each scale in the multi-scale feature map with the intermediate feature map to obtain each fused feature map; an insulator detection task head network, used to detect each fused feature map, and obtain a defect detection result of the image to be detected. Segmenting and detecting features of feature maps by different task heads can improve the real-time and accuracy of defect detection, and solve the problem of low real-time and accuracy of insulator defect detection methods in the prior art.

[0122] Furthermore, the insulator segmentation task head network 320 includes: obtaining shallow features in the multi-scale feature map; performing a convolution operation on the shallow features through the insulator segmentation task head network to obtain an intermediate feature map corresponding to the image to be detected.

[0123] Furthermore, the fusion network 330 includes: for each scale feature map in the multi-scale feature map, adjusting the scale of the intermediate feature map to a scale matching the feature map through a feature scale adaptation module; fusing the adjusted intermediate feature map with the feature map through a fusion network to obtain a fused feature map.

[0124] Furthermore, the insulator detection task head network 340 includes: for fused feature maps at different scales, respectively detecting the fused feature maps through the positioning branch and the classification branch in the insulator detection task head network, obtaining the positioning frame information of the fused feature map and the corresponding object confidence and category; taking the positioning frame in the positioning frame information that contains the object confidence that meets the requirements as the defect detection result of the image to be detected.

[0125] Furthermore, the fused feature map is detected by the positioning branch and the classification branch in the insulator detection task head network to obtain the positioning frame information of the fused feature map and the corresponding confidence and category of the contained object, including: extracting the positioning feature corresponding to the positioning branch and the classification feature corresponding to the classification branch from the fused feature map; detecting the fused feature map based on the positioning feature by the positioning branch to obtain the positioning frame information of the fused feature map and the corresponding confidence of the contained object; detecting the fused feature map based on the classification feature by the classification branch to obtain the category corresponding to the positioning frame information.

[0126] Furthermore, after obtaining the positioning frame information, the device further includes: deleting redundant positioning frames in the positioning frame information by using a non-maximum suppression algorithm.

[0127] Furthermore, the device also includes: performing multi-task joint training on the insulator segmentation task head network and the insulator detection task head network through a loss function; the loss function includes a target detection task loss function and a segmentation task loss function, and the detection task loss function includes a classification loss function, a confidence loss function and a target box loss function.

[0128] The above-mentioned insulator defect detection device can execute the insulator defect detection method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0129] Embodiment 4

[0130] Fig.11 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0131] like Fig.11 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0132] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0133] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the insulator defect detection method.

[0134] In some embodiments, the insulator defect detection method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the insulator defect detection method described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to perform the insulator defect detection method in any other appropriate manner (e.g., by means of firmware).

[0135] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0136] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0137] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0138] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0139] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0140] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0141] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0142] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting insulator defects, characterized in that: The method comprises: The feature extraction and enhancement module is used to extract and enhance the features of the image to be detected, thereby obtaining a multi-scale feature map corresponding to the image to be detected; Segmenting the image to be detected based on shallow features in the multi-scale feature map through an insulator segmentation task head network to obtain an intermediate feature map; The feature map of each scale in the multi-scale feature map is respectively fused with the intermediate feature map through a fusion network to obtain each fused feature map; The fused feature maps are detected through the insulator detection task head network to obtain the defect detection result of the image to be detected.

2. The method according to claim 1, characterized in that The step of segmenting the image to be detected based on shallow features in the multi-scale feature map through the insulator segmentation task head network to obtain an intermediate feature map includes: Obtaining shallow features in the multi-scale feature map; The shallow features are convolved through the insulator segmentation task head network to obtain an intermediate feature map corresponding to the image to be detected.

3. The method according to claim 1, characterized in that The step of fusing the feature map of each scale in the multi-scale feature map with the intermediate feature map to obtain each fused feature map through a fusion network includes: For each scale feature map in the multi-scale feature map, adjusting the scale of the intermediate feature map to a scale matching the feature map by a feature scale adaptation module; The adjusted intermediate feature map is fused with the feature map through a fusion network to obtain a fused feature map.

4. The method according to claim 1, characterized in that: The insulator detection task head network is used to detect each fused feature map to obtain a defect detection result of the image to be detected, including: For the fused feature maps at different scales, the fused feature maps are respectively detected by the positioning branch and the classification branch in the insulator detection task head network to obtain the positioning frame information of the fused feature maps and the corresponding confidence and category of the included objects; The positioning frame in the positioning frame information that contains an object confidence that meets the requirements is used as the defect detection result of the image to be detected.

5. The method according to claim 4, characterized in that The fused feature map is detected by the positioning branch and the classification branch in the insulator detection task head network to obtain the positioning frame information of the fused feature map and the corresponding confidence and category of the included object, including: Extracting the positioning feature corresponding to the positioning branch and the classification feature corresponding to the classification branch from the fused feature map; Detecting the fused feature map based on the positioning feature through the positioning branch to obtain positioning frame information of the fused feature map and corresponding confidence of the included object; The fused feature map is detected based on the classification feature by the classification branch to obtain a category corresponding to the positioning frame information.

6. The method according to claim 4, characterized in that After obtaining the positioning frame information, the method further includes: Redundant positioning frames in the positioning frame information are deleted by a non-maximum suppression algorithm.

7. The method according to claim 1, characterized in that The method further comprises: The insulator segmentation task head network and the insulator detection task head network are jointly trained with multi-tasks through loss functions; the loss function includes a target detection task loss function and a segmentation task loss function, and the detection task loss function includes a classification loss function, a confidence loss function and a target box loss function.

8. An insulator defect detection device, characterized in that: The device comprises: A feature extraction and enhancement module is used to extract and enhance features of the image to be detected to obtain a multi-scale feature map corresponding to the image to be detected; An insulator segmentation task head network is used to segment the image to be detected based on shallow features in the multi-scale feature map to obtain an intermediate feature map; A fusion network, used for fusing the feature map of each scale in the multi-scale feature map with the intermediate feature map to obtain each fused feature map; The insulator detection task head network is used to detect the fused feature maps to obtain the defect detection result of the image to be detected.

9. An electronic device, characterized in that: The device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the insulator defect detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the insulator defect detection method according to any one of claims 1 to 7 when executed.

Citation Information

Patent Citations

  • Laser point cloud 3D target detection method

    CN114627183A

  • Power distribution network insulator defect detection method and system based on context multi-output fusion module

    CN118379559A

Cited By

  • Defect detection method and detection system based on train track wear

    CN120298308A

  • Insulator damage detection method and device and electronic equipment

    CN120807518A