Insulator defect identification method, device and related equipment
Through the combination of weakly supervised recognition model and metric learning model, the problem that deep learning object detection algorithm relies on manual annotation of samples is solved, and the recognition accuracy and efficiency of insulator defect detection are improved.
Patent Information
- Application Number
- CN202411267474.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-11
- Publication Date
- 2024-12-31
- Estimated Expiration
- 2044-09-11
AI Technical Summary
The existing insulator defect detection method based on deep learning object detection algorithm relies on a large number of manual annotation samples, resulting in a reduced recognition accuracy and annotation uncertainty.
A weakly supervised recognition model is used to locate and slice the insulator area in the drone image, generate an insulator slice data set, and a metric learning model is trained based on this data set to identify insulator defects.
It improves the identification accuracy of insulator defects, reduces the dependence on manual labeling samples, and enhances the accuracy and efficiency of detection.
Smart Images
Figure CN118799763B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of defect identification, and in particular to an insulator defect identification method, device and related equipment. Background Art
[0002] Insulator defect detection plays a vital role in the power system. It not only ensures the safety of power supply, but also significantly improves the reliability of power supply by preventing power accidents caused by insulation failure, reducing line failures or power outages caused by insulator defects. In addition, regular inspections help extend the service life of insulators, reduce replacement costs, and detect minor defects in time to prevent them from developing into larger problems, thereby reducing maintenance costs. Insulator defect detection also meets the regular inspection and maintenance requirements of power industry regulations, improves the operating efficiency of the system, and ensures the optimal operating state of the power system by reducing power losses.
[0003] In the early days of the development of power systems, the detection of insulator defects mainly relied on manual observation and visual inspection using simple tools such as telescopes. With the development of technology, some detection technologies based on physical principles have begun to emerge, such as capacitor discharge method, insulation resistance measurement method, ultraviolet imaging method, infrared imaging method, etc. In recent years, deep learning technology has played an important role in insulator defect detection. For example, the practice of wire insulator defect detection based on the YOLOv5 algorithm has improved the accuracy and efficiency of detection by building data sets and model training. However, the algorithm based on deep learning target detection is a strongly supervised learning algorithm. The improvement of recognition accuracy depends on a large number of manually labeled samples, and there will be uncertainty in the labeling, which ultimately leads to a decrease in the recognition accuracy of insulator defects. Summary of the invention
[0004] Based on this, it is necessary to provide an insulator defect identification method, device and related equipment that can improve the identification accuracy of insulator defects in response to the above technical problems.
[0005] In a first aspect, the present application provides an insulator defect identification method, the method comprising:
[0006] The trained weakly supervised recognition model is used to locate the insulator area in the UAV image;
[0007] The insulator area is sliced using the trained weakly supervised recognition model to obtain an insulator slice dataset;
[0008] The metric learning model is trained based on the insulator slice data set to obtain a trained metric learning model;
[0009] The trained metric learning model is used to identify insulator defects in UAV images.
[0010] In one embodiment, the insulator region in the drone image is located using a trained weakly supervised recognition model, including:
[0011] Use the trained weakly supervised recognition model to obtain the insulator thermal map;
[0012] The insulator thermal map is segmented by adaptive threshold segmentation to obtain the insulator segmentation result map;
[0013] The insulator area is located according to the insulator segmentation result diagram.
[0014] In one embodiment, the insulator defects in the drone image are identified using the trained metric learning model, including:
[0015] The trained metric learning model is used to extract the insulator defect features and non-defect dataset features from the dataset composed of drone images respectively;
[0016] Determine the similarity between the insulator defect characteristics and the non-defect data set characteristics through a metric function;
[0017] When the similarity is greater than a preset threshold, it is determined that an insulator defect exists in the drone image.
[0018] In one embodiment, before locating the insulator region in the drone image using the trained weakly supervised recognition model, the method further includes:
[0019] The UAV insulator dataset obtained from UAV images is used to train the weakly supervised recognition model to obtain a trained weakly supervised recognition model.
[0020] In a second aspect, the present application also provides an insulator defect identification device, the device comprising:
[0021] The positioning module is used to locate the insulator area in the input UAV image using the trained weakly supervised recognition model;
[0022] A slicing module, used for slicing the insulator region to obtain an insulator slice data set;
[0023] A second training module is used to train the metric learning model based on the insulator slice data set to obtain a trained metric learning model;
[0024] The recognition module is used to identify insulator defects in the UAV image using a trained metric learning model.
[0025] In a third aspect, the present application further provides a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned insulator defect identification method when executing the computer program.
[0026] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned insulator defect identification method are implemented.
[0027] In a fifth aspect, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned insulator defect identification method are implemented.
[0028] The above-mentioned insulator defect recognition method slices the insulator area by using the trained weak supervision recognition model to obtain an insulator slice data set; trains the metric learning model based on the insulator slice data set to obtain a trained metric learning model; and uses the trained metric learning model to recognize insulator defects in drone images. The insulator defect recognition method of the present application can improve the recognition accuracy of insulator defects. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 A schematic diagram of a flow chart of an insulator defect identification method in one embodiment;
[0030] Figure 2 A flowchart of a training process for a weakly supervised recognition model in one embodiment;
[0031] Figure 3 is a structural diagram of a weakly supervised recognition model in one embodiment;
[0032] Figure 4 A structural diagram of a structural diagram of a weakly supervised recognition model in one embodiment;
[0033] Figure 5 A diagram showing a metric function network structure in one embodiment;
[0034] Figure 6 A schematic diagram of tensor voting in one embodiment;
[0035] Figure 7 is a structural block diagram of an insulator defect recognition device in a drone image in one embodiment;
[0036] Figure 8 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION
[0037] 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.
[0038] Insulator defect detection plays a vital role in the power system. It not only ensures the safety of power supply, but also significantly improves the reliability of power supply by preventing power accidents caused by insulation failure, and reduces line failures or power outages caused by insulator defects. In addition, regular inspections help extend the service life of insulators, reduce replacement costs, and detect minor defects in time to prevent them from developing into larger problems, thereby reducing maintenance costs. Insulator defect detection also meets the regular inspection and maintenance requirements of power industry regulations, improves the operating efficiency of the system, and ensures the optimal operating state of the power system by reducing power loss. In the early days of the development of power systems, the detection of insulator defects mainly relied on manual observation and visual inspection using simple tools such as telescopes. With the development of technology, some detection technologies based on physical principles have begun to emerge, such as capacitor discharge method, insulation resistance measurement method, ultraviolet imaging method, infrared imaging method, etc. In recent years, deep learning technology has played an important role in insulator defect detection. For example, the practice of wire insulator defect detection based on the YOLOv5 algorithm has improved the accuracy and efficiency of detection by building data sets and model training. However, the algorithm based on deep learning target detection is a strongly supervised learning algorithm. The improvement of recognition accuracy depends on a large number of manually labeled samples, and there will be uncertainty in the labeling, which ultimately leads to a decrease in the recognition accuracy of insulator defects. Based on this, the embodiment of the present application provides an insulator defect recognition method to improve the above technical problems.
[0039] In one embodiment, Figure 1 An insulator defect identification method is provided according to an embodiment of the present application, and the method is applied to a server as an example for explanation. The method includes the following steps:
[0040] S101, using the trained weakly supervised recognition model to locate the insulator area in the input UAV image.
[0041] Optionally, before using the trained weakly supervised recognition model to locate the insulator area in the UAV image, it is also necessary to train the weakly supervised recognition model using a UAV insulator data set obtained from the UAV image to obtain a trained weakly supervised recognition model.
[0042] Specifically, after completing the training of the weakly supervised recognition model, the UAV image is input into the weakly supervised recognition model to locate the insulator region in the UAV image.
[0043] S102, slicing the insulator region to obtain an insulator slice data set.
[0044] Optionally, slicing the insulator region refers to cutting the insulator region into a plurality of small pieces to increase the loading speed of a large image.
[0045] S103, training a metric learning model based on the insulator slice data set to obtain a trained metric learning model.
[0046] S104, using the trained metric learning model to identify insulator defects in the UAV image.
[0047] The above-mentioned insulator defect recognition method slices the insulator area by using the trained weak supervision recognition model to obtain an insulator slice data set; trains the metric learning model based on the insulator slice data set to obtain a trained metric learning model; and uses the trained metric learning model to recognize insulator defects in drone images. The insulator defect recognition method of the present application can improve the recognition accuracy of insulator defects.
[0048] Based on the above embodiments, Figure 2 The training process of the weakly supervised recognition model is explained. Figure 2 As shown, the training of the weakly supervised recognition model is based on the Transformer and reversible Transformer architecture; the drone image is input into an encoder composed of four Transformer blocks, four reversible Transformer blocks are constructed and the four reversible Transformer blocks are connected at the Transformer blocks respectively; the gradient information is transmitted to the global mean pooling layer through the encoder and the multi-layer perceptron, and the classification result is obtained as the output.
[0049] It can be understood that the present embodiment provides a possible implementation method for training the weakly supervised recognition model. The method is simple and easy to operate, and can improve the recognition accuracy of insulator defects.
[0050] Based on the above embodiments, Figure 3 The structural diagram of the weakly supervised recognition model is explained. Figure 3 As shown, after the weakly supervised recognition model is trained, the drone image is recognized, the auxiliary branch is cancelled in the feature extraction stage, and the output of the multi-layer perceptron is multiplied by the output of the global mean pooling to obtain an insulator heat map; the insulator heat map is segmented by an adaptive threshold segmentation algorithm to obtain an insulator segmentation result map.
[0051] It can be understood that the structural diagram of the weakly supervised recognition model is given in this embodiment, and the structural diagram is simple and clear and easy to understand.
[0052] Based on the above embodiments, Figure 4 The structural diagram of the metric learning model is explained. Figure 4 As shown, the metric learning model uses the ResNet50 backbone network to extract features from the support image dataset and the query image dataset to obtain support features and query features; tensor voting is used to integrate feature information from different scales, and the similarity between features is measured by a metric function; when the similarity is greater than a preset threshold, it is determined that there is an insulator defect in the drone image.
[0053] Optional, such as Figure 5 As shown in the figure, the metric function network structure includes two parts of metric network structures, namely structural feature metric and fine-grained feature metric. In the structural feature metric structure, larger convolution kernels of 7x7 and 5x5 and the maximum pooling layer are mainly used; in the fine-grained feature metric structure, 3x3 convolution kernels and the average pooling layer are mainly used.
[0054] It can be understood that the structural diagram of the metric learning model is given in this embodiment, and the structural diagram is simple and clear and easy to understand.
[0055] Based on the above embodiments, Figure 6 The schematic diagram of tensor voting is explained. Figure 6 As shown in Figure 2, tensor voting uses a log-polar voting mechanism to collect key votes from both close and far distance regions of the support and query features. Therefore, tensor voting is able to integrate feature information from different distances, thus achieving more comprehensive analysis in visual recognition tasks.
[0056] Optional, such as Figure 6 As shown in the figure, the feature point will receive votes from the feature points in its voting domain. The votes cast by the points in the voting domain are linearly accumulated to generate a new feature tensor at the point. The newly generated feature tensor incorporates the feature tensor information in the voting domain in the hope of obtaining a richer and more expressive feature representation.
[0057] It can be understood that the structural diagram of the weakly supervised recognition model is given in this embodiment, and the structural diagram is simple and clear and easy to understand.
[0058] It should be understood that, although the steps in the flowcharts involved in the above 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 is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may 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 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.
[0059] Based on the same inventive concept, the embodiment of the present application also provides an insulator defect identification device for implementing the above-mentioned insulator 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 the one or more insulator defect identification device embodiments provided below can refer to the above-mentioned limitations on an insulator defect identification method, and will not be repeated here.
[0060] In one embodiment, by Figure 7 FIG. 2 shows a structural block diagram of an insulator defect identification device in one embodiment. Figure 7 As shown, an insulator defect identification device 7 is provided, wherein the device 7 includes a positioning module 70, a slicing module 71, a first training module 72 and an identification module 73, wherein:
[0061] A positioning module 70 is used to locate the insulator region in the input UAV image using the trained weakly supervised recognition model;
[0062] A slicing module 71, used for slicing the insulator region to obtain an insulator slice data set;
[0063] A second training module 72, configured to train the metric learning model based on the insulator slice data set to obtain a trained metric learning model;
[0064] The identification module 73 is used to identify the insulator defects in the UAV image using the trained metric learning model.
[0065] The above-mentioned insulator defect recognition device in drone images slices the insulator area using a trained weakly supervised recognition model to obtain an insulator slice data set; trains a metric learning model based on the insulator slice data set to obtain a trained metric learning model; and uses the trained metric learning model to recognize insulator defects in drone images. The insulator defect recognition device in drone images of the present application can improve the recognition accuracy of insulator defects.
[0066] In one of the embodiments, the above-mentioned device 7 is also specifically used for: training the weakly supervised recognition model based on Transformer and reversible Transformer architecture; inputting the drone image into an encoder composed of four Transformer blocks, constructing four reversible Transformer blocks and connecting the four reversible Transformer blocks at the Transformer blocks respectively; transmitting the gradient information to the global mean pooling layer via the encoder and the multi-layer perceptron, and obtaining the classification result as output.
[0067] In one of the embodiments, the above-mentioned device 7 is also specifically used to: after the training of the weakly supervised recognition model is completed, recognize the drone image, cancel the auxiliary branch in the feature extraction stage, and perform a dot multiplication operation on the output of the multi-layer perceptron and the output of the global mean pooling to obtain an insulator heat map; segment the insulator heat map through an adaptive threshold segmentation algorithm to obtain an insulator segmentation result map.
[0068] In one embodiment, the device 7 is further used for:
[0069] The ResNet50 backbone network is used to extract features from the support image dataset and the query image dataset through the metric learning model to obtain support features and query features; the feature information from different scales is integrated through tensor voting, and the similarity between features is measured through the metric function; when the similarity is greater than a preset threshold, it is determined that there is an insulator defect in the drone image.
[0070] In one embodiment, the device 7 is further used for:
[0071] Tensor voting adopts a log-polar voting mechanism to collect key votes from both close and distant regions of support features and query features.
[0072] In one embodiment, the device 7 is further used for:
[0073] The metric learning model includes a structural feature metric structure and a fine-grained feature metric structure; wherein the structural feature metric structure uses a 7x7 or 5x5 convolution kernel and a maximum pooling layer, and the fine-grained feature metric structure uses a 3x3 convolution kernel and an average pooling layer.
[0074] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 8 The computer device includes a processor, a memory, a network interface and a transceiver connected through a system bus. 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 transceiver of the computer device is used to perform operations of receiving data or sending data under the control of the processor. 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 sample data. The network 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, an insulator defect identification method is implemented.
[0075] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion 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. Specifically, the 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.
[0076] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory. The principles and specific processes of implementing the various embodiments by the processor can be found in the description of an insulator defect identification method embodiment in the aforementioned embodiment, which will not be repeated here.
[0077] 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 principles and specific processes in each embodiment are implemented. Please refer to the description of an insulator defect identification method embodiment in the aforementioned embodiment, which will not be repeated here.
[0078] In one embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the principles and specific processes in each embodiment are implemented. Please refer to the description of an insulator defect identification method embodiment in the aforementioned embodiment, which will not be repeated here.
[0079] It should be noted that the information involved in this application (including but not limited to the insulator slice data set involved in the insulator defect identification process in this application, etc.) is information or data that has been fully authorized by all parties.
[0080] 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.
[0081] 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.
[0082] 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 insulator defects, characterized in that: The method comprises: The trained weakly supervised recognition model is used to locate the insulator area in the input UAV image; The training of the weakly supervised recognition model is based on Transformer and reversible Transformer architecture; Input the drone image to the encoder composed of four Transformer blocks, construct four reversible Transformer blocks and connect the four reversible Transformer blocks at the Transformer blocks respectively; The gradient information is transmitted to the global mean pooling layer via the encoder and the multi-layer perceptron to obtain the classification result as output; After the weakly supervised recognition model is trained, the drone image is recognized, the auxiliary branch is cancelled in the feature extraction stage, and the output of the multilayer perceptron is multiplied by the output of the global mean pooling to obtain the insulator heat map; Segment the insulator thermal map by using an adaptive threshold segmentation algorithm to obtain an insulator segmentation result map; Slicing the insulator region to obtain an insulator slice data set; Training a metric learning model based on the insulator slice data set to obtain a trained metric learning model; Using the trained metric learning model to identify insulator defects in the drone image; The metric learning model uses the ResNet50 backbone network to extract features from the support image dataset and the query image dataset to obtain support features and query features; Integrate feature information from different scales through tensor voting, and measure the similarity between features through metric functions; When the similarity is greater than a preset threshold, it is determined that an insulator defect exists in the drone image.
2. The insulator defect identification method according to claim 1, characterized in that: The method further comprises: Tensor voting adopts a log-polar voting mechanism to collect key votes from both close and distant regions of support features and query features.
3. The insulator defect identification method according to claim 1, characterized in that: The metric learning model includes a structural feature metric structure and a fine-grained feature metric structure; wherein the structural feature metric structure uses a 7x7 or 5x5 convolution kernel and a maximum pooling layer, and the fine-grained feature metric structure uses a 3x3 convolution kernel and an average pooling layer.
4. An insulator defect identification device, characterized in that: The device comprises: The positioning module is used to locate the insulator area in the input UAV image using the trained weakly supervised recognition model; The training of the weakly supervised recognition model is based on Transformer and reversible Transformer architecture; Input the drone image to the encoder composed of four Transformer blocks, construct four reversible Transformer blocks and connect the four reversible Transformer blocks at the Transformer blocks respectively; The gradient information is transmitted to the global mean pooling layer via the encoder and the multi-layer perceptron to obtain the classification result as output; After the weakly supervised recognition model is trained, the drone image is recognized, the auxiliary branch is cancelled in the feature extraction stage, and the output of the multilayer perceptron is multiplied by the output of the global mean pooling to obtain the insulator heat map; The insulator heat map is segmented by an adaptive threshold segmentation algorithm to obtain an insulator segmentation result map; a slicing module is used to slice the insulator area to obtain an insulator slice data set; A second training module is used to train the metric learning model based on the insulator slice data set to obtain a trained metric learning model; an identification module, for identifying insulator defects in the drone image using a trained metric learning model; The metric learning model uses the ResNet50 backbone network to extract features from the support image dataset and the query image dataset to obtain support features and query features; Integrate feature information from different scales through tensor voting, and measure the similarity between features through metric functions; When the similarity is greater than a preset threshold, it is determined that an insulator defect exists in the drone image.
5. 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 3 are implemented.
6. 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 3 are implemented.
7. 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 3 are implemented.