High-voltage equipment oil paper insulation defect identification method, system, equipment and medium

Abstract

The invention relates to the technical field of high-voltage equipment electrical variable detection, in particular to a high-voltage equipment oil paper insulation defect identification method, system and equipment and a medium, and aims to solve the technical problem of low identification accuracy of oil paper insulation defects in a transformer. The method comprises the following steps of: firstly, performing image registration and pixel fusion processing of a repeated area on images of a plurality of positions in the high-voltage equipment to obtain an internal spliced image of the high-voltage equipment; then, extracting different depth features of the spliced image in the high-voltage equipment, and performing feature cross fusion processing of bottom-to-top fusion and top-to-bottom fusion to obtain a plurality of cross fusion features; and finally, splicing and segmenting the plurality of cross fusion features to obtain an oil paper insulation defect identification result. The method is used in the detection process of the internal electrical variables of the high-voltage equipment, and can improve the identification accuracy of oil paper insulation defects and the detection accuracy of the electrical variables of the high-voltage equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical variable detection of high-voltage equipment, and specifically to a method, system, device, and medium for identifying oil-paper insulation defects in high-voltage equipment. Background Technique

[0002] The detection of oil-paper insulation defects in the electrical variable detection of high-voltage equipment is of great significance for preventing power system accidents. In actual operation, due to the metal sealing design of large transformers, it is difficult to conduct effective internal detection without disassembling them. Currently, existing detection techniques cannot directly observe internal faults of transformers, which poses a huge challenge to the safe operation of transformers.

[0003] To solve this problem, in-tank inspection robots for transformers have become an important research direction in the field of internal fault detection. These robots are equipped with high-precision sensors and imaging devices, and can penetrate deep inside for inspection without disassembling the transformer. Their application not only improves the detection efficiency but also reduces the risk of manual intervention. However, in-tank inspection robots still face challenges in accurately and quickly identifying and segmenting internal defect discharge traces of transformers when performing tasks. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, system, device, and medium for identifying oil-paper insulation defects in high-voltage equipment.

[0005] The technical solution of the present invention is as follows: A method for identifying oil-paper insulation defects in high-voltage equipment includes the following operations: S1. Obtain images of several positions inside the high-voltage equipment. After image registration respectively, perform pixel fusion processing on overlapping regions to obtain a spliced image inside the high-voltage equipment; S2. Extract different depth features of the spliced image inside the high-voltage equipment to obtain several depth features; specifically: the spliced image inside the high-voltage equipment is processed by convolution and cross-level zigzag feature extraction to obtain shallow-layer features inside the high-voltage equipment; the shallow-layer features inside the high-voltage equipment are respectively processed by cross-level zigzag feature extraction once and twice to obtain middle-layer features and middle-deep layer features inside the high-voltage equipment; the middle-deep layer features inside the high-voltage equipment are processed by cross-level zigzag feature extraction and spatial pyramid pooling to obtain deep-layer features inside the high-voltage equipment; S3. Perform feature cross-fusion processing on the deep-layer features inside the high-voltage equipment with the middle-deep layer features, middle-layer features, and shallow-layer features inside the high-voltage equipment to obtain the first cross-fusion feature, the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature; S4. After the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature are respectively processed by multi-scale feature enhancement and channel dimension enhancement, they are concatenated with the first cross-fusion feature to obtain the feature map to be detected; the feature map to be detected is processed by segmentation to obtain the oil-paper insulation defect recognition result.

[0006] The operation of feature cross-fusion processing in S3 is specifically as follows: The deep features inside the high-voltage equipment, the mid-deep features inside the high-voltage equipment, the middle features inside the high-voltage equipment, and the shallow features inside the high-voltage equipment are respectively processed by local feature extraction to obtain deep local features, mid-deep local features, middle local features, and shallow local features; the deep local features are respectively subjected to local feature fusion processing based on upsampling with the mid-deep local features, the middle local features, and the shallow local features to obtain the first fusion feature, the second fusion feature, and the third fusion feature; the first fusion feature is processed by multi-scale feature aggregation to obtain the first cross-fusion feature; the first cross-fusion feature is respectively subjected to local attention feature aggregation processing based on downsampling with the second fusion feature, the third fusion feature, and the deep local features to obtain the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature.

[0007] The operation of local feature extraction is specifically as follows: The input feature map is respectively processed by local attention mechanisms in the horizontal and vertical directions and then subjected to element-wise multiplication with the input feature map to obtain the local attention feature map; the local attention feature map and the input feature map are subjected to two-dimensional convolution after element-wise multiplication to obtain the output feature map; the input feature map is the deep features inside the high-voltage equipment, or the mid-deep features inside the high-voltage equipment, or the middle features inside the high-voltage equipment, or the shallow features inside the high-voltage equipment; the output feature map is the deep local features, or the mid-deep local features, or the middle local features, or the shallow local features.

[0008] The operation of local feature fusion processing based on upsampling is specifically as follows: The deep local features are processed by several transposed convolutions to obtain deep transposed convolution local features; the deep transposed convolution local features are respectively processed by local attention mechanisms in the horizontal and vertical directions and then subjected to element-wise multiplication with the deep transposed convolution local features and the first feature to be fused to obtain the initial first fusion feature; the initial first fusion feature and the deep transposed convolution local features are subjected to element-wise addition to obtain the fusion feature; the first feature to be fused is the mid-deep local features, or the middle local features, or the shallow local features.

[0009] The operation of local attention feature aggregation processing based on downsampling is specifically as follows: The first cross-fusion feature is processed by two-dimensional convolution several times to obtain the first cross-fusion convolution feature; after the first cross-fusion convolution feature is processed by the local attention mechanism in the horizontal and vertical directions respectively, it is multiplied element by element with the first cross-fusion convolution feature and the second feature to be fused to obtain the initial second fusion feature; the initial second fusion feature and the first cross-fusion convolution feature are processed by element-wise addition and multi-scale feature aggregation processing to obtain the cross-fusion feature; the second feature to be fused is the second fusion feature, or the third fusion feature, or the deep local feature.

[0010] Before the operation of multiplying the second feature to be fused element by element, it also includes performing local feature enhancement processing on the second feature to be fused to obtain the second enhanced feature to be fused; when the second feature to be fused is the second fusion feature or the third fusion feature, the operation of local feature enhancement processing can be achieved through multi-scale feature aggregation processing, local attention mechanism processing and two-dimensional convolution processing; when the second feature to be fused is the deep local feature, the operation of local feature enhancement processing can be achieved through local attention mechanism processing and two-dimensional convolution processing.

[0011] After the operation of obtaining the internal splicing diagram of the high-voltage equipment in S1, it also includes performing image enhancement processing on the internal splicing diagram of the high-voltage equipment to obtain the internal enhanced diagram of the high-voltage equipment for performing the operation in S2; the operation of image enhancement processing is specifically as follows: the internal splicing diagram of the high-voltage equipment is processed by Gaussian filtering at different scales to obtain several Gaussian filtering feature maps; after several Gaussian filtering feature maps and the internal splicing diagram of the high-voltage equipment are processed by feature aggregation, they are normalized to obtain the internal enhanced diagram of the high-voltage equipment.

[0012] A high-voltage equipment oil-paper insulation defect recognition system for implementing the above-mentioned high-voltage equipment oil-paper insulation defect recognition method includes: An internal splicing diagram generation module of the high-voltage equipment, which is used to obtain images at several positions inside the high-voltage equipment, and after image registration respectively, perform pixel fusion processing on the overlapping areas to obtain the internal splicing diagram of the high-voltage equipment; Several deep feature generation modules, which are used to extract different deep features of the internal splicing diagram of the high-voltage equipment to obtain several deep features; specifically: the internal splicing diagram of the high-voltage equipment is processed by convolution and cross-level zigzag feature extraction to obtain the internal shallow feature of the high-voltage equipment; the internal shallow feature of the high-voltage equipment is processed by cross-level zigzag feature extraction once and twice respectively to obtain the internal middle feature and the internal middle-deep feature of the high-voltage equipment; the internal middle-deep feature of the high-voltage equipment is processed by cross-level zigzag feature extraction and spatial pyramid pooling to obtain the internal deep feature of the high-voltage equipment; The cross - fusion feature generation module is used to perform feature cross - fusion processing on the deep features inside the high - voltage equipment, the mid - deep features inside the high - voltage equipment, the middle - layer features inside the high - voltage equipment, and the shallow - layer features inside the high - voltage equipment, to obtain the first cross - fusion feature, the second cross - fusion feature, the third cross - fusion feature, and the fourth cross - fusion feature; The oil - paper insulation defect recognition result generation module is used to splice the second cross - fusion feature, the third cross - fusion feature, and the fourth cross - fusion feature with the first cross - fusion feature after they are respectively processed by multi - scale feature enhancement and channel - dimension enhancement, to obtain the to - be - detected feature map; the to - be - detected feature map is processed by segmentation to obtain the oil - paper insulation defect recognition result.

[0013] A high - voltage equipment oil - paper insulation defect recognition device includes a processor and a memory. Among them, when the processor executes the computer program stored in the memory, it implements the above - mentioned high - voltage equipment oil - paper insulation defect recognition method.

[0014] A computer - readable storage medium is used to store a computer program. Among them, when the computer program is executed by the processor, it implements the above - mentioned high - voltage equipment oil - paper insulation defect recognition method.

[0015] The beneficial effects of the present invention are as follows: A high - voltage equipment oil - paper insulation defect recognition method provided by the present invention first obtains a spliced map inside the high - voltage equipment by performing image registration and pixel fusion processing on the repeated areas of the images obtained at several positions inside the high - voltage equipment; then, extracts different - depth features of the spliced map inside the high - voltage equipment, and performs feature cross - fusion processing of bottom - up fusion and top - down fusion on the different - depth features of the spliced map inside the high - voltage equipment, so that the image can not only focus on local details but also combine overall semantic information, improve the feature expression ability of the carbon - trace image, and obtain the first cross - fusion feature, the second cross - fusion feature, the third cross - fusion feature, and the fourth cross - fusion feature, which respectively reflect different - level fusion feature information; finally, the second cross - fusion feature, the third cross - fusion feature, and the fourth cross - fusion feature are respectively processed by multi - scale feature enhancement and channel - dimension enhancement and then spliced with the first cross - fusion feature to enhance the image aggregation dynamics and expressiveness, realize the separation of the oil - paper insulation carbon - trace defect from the complex environmental background, obtain the to - be - detected feature map; the to - be - detected feature map is processed by segmentation to obtain the oil - paper insulation defect recognition result; this method is used for the recognition of oil - paper insulation defects in the process of detecting electrical variables inside high - voltage equipment, and can improve the accuracy of defect recognition and the accuracy of detecting electrical variables of high - voltage equipment. Description of the Drawings

[0016] The solutions and advantages of the present application will become clear to those of ordinary skill in the art by reading the following detailed description of the preferred embodiments. The drawings are only for the purpose of illustrating the preferred embodiments and are not considered to be a limitation of the present invention.

[0017] In the drawings: Figure 1 It is a schematic diagram of the effect of image enhancement processing in the embodiment. Figure 1 In (a) is the original image of carbon marks on oil-paper insulation. Figure 1 In (b) is the image of carbon marks on oil-paper insulation after image enhancement. Figure 2 It is a schematic diagram of the result of segmentation processing in the embodiment. Figure 2 In (a) is the recognition result map of dendritic carbon marks. Figure 2 In (b) is the recognition result map of clustered carbon marks. Specific Embodiments

[0018] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the drawings.

[0019] This embodiment provides a method for identifying oil-paper insulation defects in high-voltage equipment, including the following operations: S1. Obtain images at several positions inside the high-voltage equipment. After image registration respectively, perform pixel fusion processing on the overlapping regions to obtain a spliced image inside the high-voltage equipment. S2. Extract different-depth features of the spliced image inside the high-voltage equipment to obtain several depth features. Specifically: the spliced image inside the high-voltage equipment undergoes convolution and cross-level zigzag feature extraction processing to obtain shallow features inside the high-voltage equipment; the shallow features inside the high-voltage equipment respectively undergo one-time and two-time cross-level zigzag feature extraction processing to obtain middle features and medium-deep features inside the high-voltage equipment; the medium-deep features inside the high-voltage equipment undergo cross-level zigzag feature extraction and spatial pyramid pooling processing to obtain deep features inside the high-voltage equipment. S3. Perform feature cross-fusion processing on the deep features inside the high-voltage equipment with the medium-deep features, middle features, and shallow features inside the high-voltage equipment to obtain the first cross-fusion feature, the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature. S4. After the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature respectively undergo multi-scale feature enhancement and channel dimension enhancement processing, splice them with the first cross-fusion feature to obtain a feature map to be detected. The feature map to be detected undergoes segmentation processing to obtain the recognition result of oil-paper insulation defects.

[0020] S1. Obtain images at several positions inside the high-voltage equipment. After image registration respectively, perform pixel fusion processing on the overlapping regions to obtain a spliced image inside the high-voltage equipment.

[0021] By performing image registration and pixel fusion processing of the repeated regions on the images obtained from several positions inside the high-voltage equipment, the image acquisition and stitching of the components inside the high-voltage equipment with a close distance and wide field of view are realized, and a stitched image of the inside of the high-voltage equipment is obtained, which is convenient for comprehensively analyzing the inside of the high-voltage equipment and is beneficial to improving the accuracy of identifying the overall oil-paper insulation defects.

[0022] First, images of several positions inside the high-voltage equipment are obtained as the data to be inspected. Specifically, several cameras are installed around the submersible in oil. When the submersible in oil navigates inside the high-voltage equipment, the several cameras simultaneously capture image videos of the inside of the high-voltage equipment in multiple directions including but not limited to the upper, lower, left, right, front, and rear directions, for extracting images of several positions inside the high-voltage equipment to improve the efficiency of obtaining the oil-paper diagrams. The captured image videos are transmitted to an industrial control computer outside the high-voltage equipment through wireless communication for display, processing, and storage, and the number of the above cameras is not less than 6.

[0023] Then, during the shooting process of the cameras around the submersible in oil, due to the relatively narrow space inside the high-voltage equipment, the shooting distance of a single camera of the submersible in oil is relatively close and the field of view is small. Identifying the oil-paper insulation defects only based on one or two images may affect the accuracy of identifying the overall oil-paper insulation defects. To solve this technical problem in this embodiment, after the images of several positions inside the high-voltage equipment are respectively subjected to image registration, pixel fusion processing of the repeated regions is performed to obtain a stitched image of the inside of the high-voltage equipment that is convenient for comprehensively analyzing the inside of the high-voltage equipment and improving the accuracy of identifying the overall oil-paper insulation defects. The above image registration can be realized by the Scale-Invariant Feature Transform method - SIFT method.

[0024] Among them, when the camera shoots images, there will be repeated regions on two adjacent images. To improve the stitching effect and the quality of the stitched image of the inside of the high-voltage equipment, in this embodiment, pixel fusion processing of the repeated regions of two adjacent images is performed to make the transition at the stitching place natural and convenient for improving the subsequent identification accuracy.

[0025] The operation of the pixel fusion processing of the repeated regions is realized through the following formula: , , is the pixel value at the corresponding position with coordinates x , y in the repeated region of the stitched image, is the pixel mixing weight of the previous image (after registration), is the coordinate of the repeated region in the previous image (after registration) with coordinates x ,y The pixel value at the corresponding position is the pixel mixing weight of the current image (after registration). is the coordinate at the repeated area in the current image (after registration) as x 、 y The pixel value at the corresponding position is the i th image's pixel mixing weight is the width of the standard image The height of the standard image 、 are respectively the i th image's width and height

[0026] To solve the problems of image exposure and color distortion inside high-voltage equipment and improve the subsequent recognition accuracy, after obtaining the internal mosaic image of high-voltage equipment in S1, it further includes performing image enhancement processing on the internal mosaic image of high-voltage equipment. Refer to Figure 1 to obtain the internal enhanced image of high-voltage equipment for performing the operations in S2

[0027] The operations of the image enhancement processing are specifically as follows: The internal mosaic image of high-voltage equipment is processed by Gaussian filtering at different scales to obtain several Gaussian filtering feature maps; after the several Gaussian filtering feature maps and the internal mosaic image of high-voltage equipment are processed by feature aggregation and then normalized, the internal enhanced image of high-voltage equipment is obtained

[0028] The feature aggregation processing and the normalization processing are respectively implemented by the following formulas , , is the pixel value at the coordinate x 、 y in the internal feature aggregation map of high-voltage equipment at the corresponding position is the n th scale's Gaussian filtering feature map's corresponding weight N is the total number of times of Gaussian filtering is the pixel value at the coordinate x 、 y in the internal mosaic image of high-voltage equipment at the corresponding position is the pixel value at the coordinate x 、 y in the nth scale's Gaussian filtering feature map at the corresponding position is the pixel value at the coordinate x 、 y in the internal enhanced image of high-voltage equipment at the corresponding position is the coordinate in several Gaussian filtering feature mapsx , y The maximum pixel value at the corresponding position, is the minimum pixel value at the corresponding position in several Gaussian-filtered feature maps with coordinates x and y .

[0029] S2. Extract different-depth features of the internal splicing map of the high-voltage equipment to obtain several depth features.

[0030] Extracting different-depth features of the internal splicing map of the high-voltage equipment can obtain rich feature information, which is beneficial to improving the subsequent ability to capture slender and curved local structures in the internal splicing map of the high-voltage equipment, facilitating the capture of dendritic structure features on the oil paper, and realizing the accurate extraction of fine dendritic structure features of carbon trace defects on the oil paper, thereby achieving higher recognition accuracy.

[0031] Specifically: The internal splicing map of the high-voltage equipment undergoes convolution and cross-level zigzag feature extraction processing to obtain the texture change features of the oil paper and obtain the internal shallow-layer features of the high-voltage equipment; the internal shallow-layer features of the high-voltage equipment undergo cross-level zigzag feature extraction processing once and twice respectively to obtain the internal middle-layer features and internal middle-deep features of the high-voltage equipment that can respectively reflect details such as micro-cracks and local color differences of the oil paper; the internal middle-deep features of the high-voltage equipment undergo cross-level zigzag feature extraction and spatial pyramid pooling processing (preferably implemented through the SPPF model) to obtain the internal deep-layer features of the high-voltage equipment that can reflect the overall structural state of the oil paper.

[0032] The operation of cross-level zigzag feature extraction processing can be implemented through the C2f module that replaces convolution with dynamic snake convolution. Specifically, the key convolution in the C2f module in the YOLO model is replaced with dynamic snake convolution - DySnakeConv convolution to obtain an improved C2f module, and the operation of cross-level zigzag feature extraction processing can be implemented through the improved C2f module.

[0033] S3. Perform feature cross-fusion processing on the internal deep-layer features of the high-voltage equipment, the internal middle-deep features of the high-voltage equipment, the internal middle-layer features of the high-voltage equipment, and the internal shallow-layer features of the high-voltage equipment to obtain the first cross-fusion feature, the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature.

[0034] Perform feature cross-fusion processing on the different-depth features of the internal splicing map of the high-voltage equipment through bottom-up fusion and top-down fusion, so that the obtained image can not only focus on local details but also combine overall semantic information, improve the feature expression ability of the carbon trace image, and obtain the first cross-fusion feature, the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature, which respectively reflect different levels of fusion feature information and are beneficial to subsequent accurate and comprehensive detection of oil-paper insulation defects.

[0035] The operation of feature cross - fusion processing is as follows.

[0036] First, the deep - layer features inside the high - voltage equipment, the mid - deep - layer features inside the high - voltage equipment, the mid - layer features inside the high - voltage equipment, and the shallow - layer features inside the high - voltage equipment are respectively processed by local feature extraction to obtain deep - layer local features, mid - deep - layer local features, mid - layer local features, and shallow - layer local features.

[0037] The operation of local feature extraction processing is specifically as follows: After the input feature map is processed by local attention mechanisms in the horizontal and vertical directions respectively, it is multiplied element - by - element with the input feature map to focus on key details, obtaining a local attention feature map that can pay attention to small targets on the oil paper, which is convenient for more clearly capturing minute defect features in the follow - up; the local attention feature map and the input feature map are multiplied element - by - element and then subjected to two - dimensional convolution processing to further enhance local features, obtaining an output feature map.

[0038] The input feature map is the deep - layer features inside the high - voltage equipment, or the mid - deep - layer features inside the high - voltage equipment, or the mid - layer features inside the high - voltage equipment, or the shallow - layer features inside the high - voltage equipment; the output feature map corresponds to the deep - layer local features, or the mid - deep - layer local features, or the mid - layer local features, or the shallow - layer local features.

[0039] Then, the deep - layer local features are respectively subjected to local feature fusion processing based on upsampling with the mid - deep - layer local features, mid - layer local features, and shallow - layer local features to achieve complementary refinement of features at different levels, obtaining the first fusion feature, the second fusion feature, and the third fusion feature, which are used for subsequent defect recognition to accurately locate and judge the degree and scope of aging defects, facilitating the analysis of the influence scope of discharge and the specific changes to local materials, so as to more accurately judge whether the oil paper is in the aging process or the moisture - affected state through detail differences, thereby improving the accuracy and stability of identifying oil - paper insulation defects in complex environments.

[0040] The operation of the above - mentioned local feature fusion processing based on upsampling is specifically as follows: The deep - layer local features are subjected to several inverse convolution processes to obtain deep - layer inverse - convolution local features; after the deep - layer inverse - convolution local features are processed by local attention mechanisms in the horizontal and vertical directions respectively, they are multiplied element - by - element with the deep - layer inverse - convolution local features and the first feature to be fused to obtain the initial first fusion feature; the initial first fusion feature and the deep - layer inverse - convolution local features are added element - by - element to obtain the fusion feature. The initial first feature to be fused is the mid - deep - layer local features, or the mid - layer local features, or the shallow - layer local features. The fusion features respectively correspond to the third fusion feature, or the second fusion feature, or the first fusion feature.

[0041] Among them, the number of times the deep local features are processed by transposed convolution in the process of obtaining the first fusion feature is 3, and the number of times the deep local features are processed by transposed convolution in the process of obtaining the second fusion feature is 2.

[0042] Finally, the first fusion feature is processed by multi-scale feature aggregation (which can be implemented by the C2f module) to obtain the first cross-fusion feature; the first cross-fusion feature is respectively processed by local attention feature aggregation based on downsampling with the second fusion feature, the third fusion feature, and the deep local features to further fuse features at different levels and enhance the image detail expression ability, obtaining the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature.

[0043] The operation of local attention feature aggregation based on downsampling is specifically as follows: the first cross-fusion feature is processed by several two-dimensional convolutions to obtain the first cross-fusion convolution feature; after the first cross-fusion convolution feature is processed by local attention mechanisms in the horizontal and vertical directions respectively, it is element-wise multiplied with the first cross-fusion convolution feature and the second feature to be fused to obtain the initial second fusion feature; the initial second fusion feature and the first cross-fusion convolution feature are processed by element-wise addition and multi-scale feature aggregation to obtain the cross-fusion feature; the second feature to be fused is the second fusion feature, or the third fusion feature, or the deep local feature; the cross-fusion features respectively correspond to the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature.

[0044] To further enhance the image detail expression ability, before the operation of element-wise multiplication of the second feature to be fused, it also includes performing local feature enhancement processing on the second feature to be fused to obtain the second enhanced feature to be fused, which is used for element-wise multiplication with the first cross-fusion convolution feature and the first cross-fusion convolution feature processed by the local attention mechanism.

[0045] When the second feature to be fused is the second fusion feature or the third fusion feature, the operation of local feature enhancement processing can be implemented by multi-scale feature aggregation processing, local attention mechanism processing, and two-dimensional convolution processing.

[0046] When the second feature to be fused is the deep local feature, the operation of local feature enhancement processing can be implemented by local attention mechanism processing and two-dimensional convolution processing.

[0047] S4. After the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature are respectively processed by multi-scale feature enhancement and channel dimension enhancement, they are concatenated with the first cross-fusion feature to obtain the feature map to be detected; the feature map to be detected is processed by segmentation to obtain the oil-paper insulation defect recognition result.

[0048] After the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature obtained through bottom-up fusion and top-down fusion are respectively processed by multi-scale feature enhancement and channel dimension enhancement, they are concatenated with the first cross-fusion feature to enhance the image aggregation dynamics and expressiveness, and realize the separation of the oil-paper insulation carbon trace defect from the complex environmental background, obtaining a to-be-detected feature map that can clearly reflect the overall state of the graph and highlight key details; the to-be-detected feature map is processed by segmentation to obtain the oil-paper insulation defect recognition result, see Figure 2 . The above channel dimension enhancement processing includes but is not limited to being implemented through the Reshape function; the operation of the segmentation processing can be implemented by a trained neural network obtained by training a neural network with a standard oil-paper insulation dataset.

[0049] Taking the second cross-fusion feature as an example, the operation of the above multi-scale feature enhancement processing is specifically as follows: the second cross-fusion feature is processed by convolution, batch normalization, Silu activation function processing, and two-dimensional convolution at different scales for several times, and then concatenated to obtain the second cross-fusion enhanced feature, which is used to execute the operation of channel dimension enhancement processing.

[0050] This embodiment also provides a high-voltage equipment oil-paper insulation defect recognition system for implementing the above high-voltage equipment oil-paper insulation defect recognition method, including: A high-voltage equipment internal splicing diagram generation module, configured to obtain images at several positions inside the high-voltage equipment, and after image registration respectively, perform pixel fusion processing on overlapping regions to obtain a high-voltage equipment internal splicing diagram; Several depth feature generation modules, configured to extract different depth features of the high-voltage equipment internal splicing diagram to obtain several depth features; specifically: the high-voltage equipment internal splicing diagram is processed by convolution and cross-level tortuous feature extraction to obtain the high-voltage equipment internal shallow layer features; the high-voltage equipment internal shallow layer features are respectively processed by cross-level tortuous feature extraction once and twice to obtain the high-voltage equipment internal middle layer features and the high-voltage equipment internal middle and deep layer features; the high-voltage equipment internal middle and deep layer features are processed by cross-level tortuous feature extraction and spatial pyramid pooling to obtain the high-voltage equipment internal deep layer features; A cross-fusion feature generation module, configured to perform feature cross-fusion processing on the high-voltage equipment internal deep layer features, the high-voltage equipment internal middle and deep layer features, the high-voltage equipment internal middle layer features, and the high-voltage equipment internal shallow layer features to obtain the first cross-fusion feature, the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature; An oil-paper insulation defect recognition result generation module, configured to respectively process the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature by multi-scale feature enhancement and channel dimension enhancement, and then concatenate them with the first cross-fusion feature to obtain a to-be-detected feature map; the to-be-detected feature map is processed by segmentation to obtain the oil-paper insulation defect recognition result.

[0051] This embodiment also provides an oil-paper insulation defect identification device for high-voltage equipment, including a processor and a memory. When the processor executes the computer program stored in the memory, the above-mentioned oil-paper insulation defect identification method for high-voltage equipment is implemented.

[0052] This embodiment also provides a computer-readable storage medium for storing a computer program. When the computer program is executed by a processor, the above-mentioned oil-paper insulation defect identification method for high-voltage equipment is implemented.

[0053] An oil-paper insulation defect identification method provided in this embodiment first performs image registration and pixel fusion processing of repeated regions on the images obtained at several positions inside the high-voltage equipment to obtain a spliced image inside the high-voltage equipment. Then, different depth features of the spliced image inside the high-voltage equipment are extracted, and the different depth features of the spliced image inside the high-voltage equipment are subjected to feature cross-fusion processing of bottom-up fusion and top-down fusion, so that the image can not only focus on local details but also combine overall semantic information, improving the feature expression ability of the carbon trace image, and obtaining the first cross-fusion feature, the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature, which respectively reflect different levels of fusion feature information. Finally, the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature are respectively subjected to multi-scale feature enhancement and channel dimension enhancement processing and then spliced with the first cross-fusion feature to enhance the image aggregation dynamics and expressiveness, realizing the separation of the oil-paper insulation carbon trace defect from the complex environmental background to obtain a to-be-detected feature map. The to-be-detected feature map is subjected to segmentation processing to obtain the oil-paper insulation defect identification result. This method is used for the identification of oil-paper insulation defects in the process of detecting electrical variables inside high-voltage equipment, and can improve the accuracy of defect identification and the accuracy of electrical variable detection of high-voltage equipment.

Claims

1. A method for identifying oil-paper insulation defects of high-voltage equipment, characterized in that: The following operations are included: S1, acquiring images of several positions inside the high-voltage equipment, performing pixel fusion processing on repeated areas after image registration, and obtaining a mosaic image of the interior of the high-voltage equipment; S2. Extract different depth features of the internal mosaic of the high-voltage equipment to obtain several depth features; specifically: the internal mosaic of the high-voltage equipment is processed by convolution and cross-level tortuous feature extraction to obtain shallow features of the internal high-voltage equipment; the shallow features of the internal high-voltage equipment are processed by cross-level tortuous feature extraction once and twice to obtain the middle features of the internal high-voltage equipment and the middle-deep features of the internal high-voltage equipment; the middle-deep features of the internal high-voltage equipment are processed by cross-level tortuous feature extraction and spatial pyramid pooling to obtain the deep features of the internal high-voltage equipment; S3, the deep features inside the high-voltage equipment are cross-fused with the medium-deep features inside the high-voltage equipment, the medium-layer features inside the high-voltage equipment, and the shallow features inside the high-voltage equipment to obtain the first cross-fusion feature, the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature; S4, the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature are respectively processed by multi-scale feature enhancement and channel dimension enhancement, and then spliced ​​with the first cross-fusion feature to obtain the feature map to be detected; the feature map to be detected is segmented to obtain the oil-paper insulation defect recognition result.

2. The method for identifying oil-paper insulation defects of high-voltage equipment according to claim 1, characterized in that: In S3, the operation of feature cross fusion processing is specifically as follows: The deep features inside the high-voltage equipment and the medium-deep features inside the high-voltage equipment, the medium-layer features inside the high-voltage equipment, and the shallow features inside the high-voltage equipment are respectively processed by local feature extraction to obtain deep local features, medium-deep local features, medium-layer local features, and shallow local features; the deep local features are respectively processed by local feature fusion based on upsampling with the medium-deep local features, the medium-layer local features, and the shallow local features to obtain the first fusion feature, the second fusion feature, and the third fusion feature; the first fusion feature is processed by multi-scale feature aggregation to obtain the first cross-fusion feature; the first cross-fusion feature is respectively processed by local attention feature aggregation based on downsampling with the second fusion feature, the third fusion feature and the deep local feature to obtain the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature.

3. The method for identifying oil-paper insulation defects of high-voltage equipment according to claim 2, characterized in that: The specific operations of local feature extraction processing are: The input feature map is processed by the local attention mechanism in the horizontal and vertical directions respectively, and then multiplied element-by-element with the input feature map to obtain the local attention feature map; the local attention feature map and the input feature map are multiplied element-by-element and then subjected to two-dimensional convolution to obtain the output feature map; The input feature map is the deep features inside the high-voltage equipment, or the medium-deep features inside the high-voltage equipment, or the medium-layer features inside the high-voltage equipment, or the shallow features inside the high-voltage equipment; the output feature map is the deep local features, or the medium-deep local features, or the medium-layer local features, or the shallow local features.

4. The method for identifying oil-paper insulation defects of high-voltage equipment according to claim 2, characterized in that: The specific operations of local feature fusion processing based on upsampling are: The deep local features are processed by several times of deconvolution to obtain the deep deconvolution local features; the deep deconvolution local features are processed by the local attention mechanism in the horizontal direction and the vertical direction respectively, and then multiplied element by element with the deep deconvolution local features and the first feature to be fused to obtain the initial first fused feature; the initial first fused feature and the deep deconvolution local features are added element by element to obtain the fused feature; The first feature to be fused is a medium-deep local feature, a medium-layer local feature, or a shallow local feature.

5. The method for identifying oil-paper insulation defects of high-voltage equipment according to claim 2, characterized in that: The specific operations of local attention feature aggregation processing based on downsampling are: The first cross-fusion feature is processed by several two-dimensional convolutions to obtain the first cross-fusion convolution feature; the first cross-fusion convolution feature is processed by the local attention mechanism in the horizontal direction and the vertical direction respectively, and then multiplied element by element with the first cross-fusion convolution feature and the second feature to be fused to obtain the initial second fusion feature; the initial second fusion feature and the first cross-fusion convolution feature are processed by element-by-element addition and multi-scale feature aggregation to obtain the cross-fusion feature; The second feature to be fused is the second fused feature, or the third fused feature, or the deep local feature.

6. The method for identifying oil-paper insulation defects of high-voltage equipment according to claim 5, characterized in that: Before the second feature to be fused is subjected to the element-by-element multiplication process, the second feature to be fused is also subjected to local feature enhancement process to obtain the second enhanced feature to be fused; When the second feature to be fused is the second fused feature or the third fused feature, the operation of local feature enhancement processing can be implemented through multi-scale feature aggregation processing, local attention mechanism processing and two-dimensional convolution processing; When the second feature to be fused is a deep local feature, the operation of local feature enhancement processing can be achieved through local attention mechanism processing and two-dimensional convolution processing.

7. The method for identifying oil-paper insulation defects of high-voltage equipment according to claim 1, characterized in that: After the operation of obtaining the internal mosaic image of the high-voltage device in S1, it also includes performing image enhancement processing on the internal mosaic image of the high-voltage device to obtain an enhanced internal image of the high-voltage device for performing the operation in S2; The specific operation of image enhancement processing is as follows: the internal splicing image of the high-voltage equipment is processed by Gaussian filtering of different scales to obtain several Gaussian filtering feature images; several Gaussian filtering feature images and the internal splicing image of the high-voltage equipment are subjected to feature aggregation processing and normalization processing to obtain the internal enhanced image of the high-voltage equipment.

8. A high voltage equipment oil-paper insulation defect identification system, characterized in that: The method for identifying oil-paper insulation defects of high-voltage equipment according to claim 1 is characterized by comprising: A high-voltage equipment internal mosaic image generation module is used to obtain images of several positions inside the high-voltage equipment, and after image registration, pixel fusion processing of repeated areas is performed to obtain a mosaic image of the internal part of the high-voltage equipment; Several deep feature generation modules are used to extract different deep features of the internal mosaic of high-voltage equipment to obtain several deep features; specifically: the internal mosaic of high-voltage equipment is processed by convolution and cross-level tortuous feature extraction to obtain shallow features of the internal high-voltage equipment; the shallow features of the internal high-voltage equipment are processed by cross-level tortuous feature extraction once and twice to obtain the middle features of the internal high-voltage equipment and the middle-deep features of the internal high-voltage equipment; the middle-deep features of the internal high-voltage equipment are processed by cross-level tortuous feature extraction and spatial pyramid pooling to obtain the deep features of the internal high-voltage equipment; A cross-fusion feature generation module is used to perform feature cross-fusion processing on the deep features inside the high-voltage equipment, the medium-deep features inside the high-voltage equipment, the medium-layer features inside the high-voltage equipment, and the shallow features inside the high-voltage equipment to obtain the first cross-fusion feature, the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature; The oil-paper insulation defect recognition result generation module is used to obtain the feature map to be detected after the second cross-fusion feature, the third cross-fusion feature, and the fourth cross-fusion feature are processed by multi-scale feature enhancement and channel dimension enhancement respectively; the feature map to be detected is segmented to obtain the oil-paper insulation defect recognition result.

9. A high voltage equipment oil-paper insulation defect identification device, characterized in that: It comprises a processor and a memory, wherein when the processor executes the computer program stored in the memory, the method for identifying oil-paper insulation defects of high-voltage equipment according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein when the computer program is executed by a processor, the method for identifying oil-paper insulation defects of high-voltage equipment according to any one of claims 1 to 7 is implemented.

Images