A method, system and storage medium for identifying defects in power equipment

Abstract

This invention relates to the field of defect identification technology and discloses a method, system, and storage medium for identifying defects in power equipment. The method includes: acquiring an image to be identified and extracting feature information from the image to generate a feature image; acquiring an original image containing equipment defects and annotating it with data to obtain a comparison image containing equipment defects, and constructing a region generation network model based on the comparison image; constructing a defect identification model using target candidate images and background candidate images, and inputting the image to be identified into the defect identification model to obtain a series of pre-selected defect boxes, determining the pre-selected defect boxes that meet the conditions; calculating the defect value of the pre-selected defect boxes that meet the conditions, and determining that the image to be identified has a defect when there is a pre-selected defect box in the image with a defect value greater than a preset threshold. This invention solves the problem of low identification accuracy in existing power equipment defect identification methods.

Description

Technical Field

[0001] This invention relates to the field of defect identification technology, and in particular to a method, system and storage medium for identifying defects in power equipment. Background Technology

[0002] With the rapid development of power technology, the application of power substation equipment is becoming increasingly widespread. Since the stability of power substation equipment plays a crucial role in the normal operation of its internal machinery, it is essential to promptly monitor and inspect the equipment to detect and resolve defects as early as possible. Currently, robot inspection is commonly used for defect inspection of power substation equipment. However, robot inspection is susceptible to interference from background and scene images due to the type of power substation equipment and the surrounding environment, leading to inaccurate defect identification. Therefore, existing methods for identifying power substation defects suffer from relatively low accuracy. Summary of the Invention

[0003] This invention provides a method, system, and storage medium for identifying defects in power equipment, in order to solve the problem of low identification accuracy in existing methods for identifying defects in power equipment.

[0004] To achieve the above objectives, the present invention employs the following technical solution:

[0005] In a first aspect, the present invention provides a method for identifying defects in power equipment, comprising:

[0006] Acquire the image to be identified, and extract the feature information from the image to be identified to generate a feature image;

[0007] The original image containing the equipment defect is acquired and labeled to obtain a comparison image containing the equipment defect. A region generation network model is then constructed based on the comparison image.

[0008] The feature image is divided into target candidate regions and background candidate regions using the region generation network model.

[0009] The target candidate region is enhanced to obtain a target candidate map, and the background candidate region is reduced to obtain a background candidate map.

[0010] A defect recognition model is constructed by using target candidate images and background candidate images. The image to be recognized is then input into the defect recognition model to obtain a series of defect pre-selection boxes, and the defect pre-selection boxes that meet the conditions are determined.

[0011] For the pre-selected defect boxes that meet the conditions, the defect value is calculated. When there is a pre-selected defect box in the image to be identified with a defect value greater than the preset threshold, the image to be identified is determined to have a defect. When there is no pre-selected defect box in the image to be identified with a defect value less than the preset threshold, the image to be identified is determined to have no defect.

[0012] Optionally, the region generation network model based on the contrast image includes:

[0013] The comparison image is input into the region generation network to construct a region generation network model.

[0014] Optionally, the step of dividing the feature image into target candidate regions and background candidate regions through the region generation network model includes:

[0015] The feature image is divided using the region generation network model to obtain pre-selected candidate regions;

[0016] The pre-selected candidate region is overlapped with the comparison image to obtain the overlap area, and the area of ​​the pre-selected candidate region is added to the area of ​​the comparison image to obtain the judgment area;

[0017] The overlapping area is compared with the determined area to obtain a determination ratio, and the determination ratio is compared with a preset determination threshold.

[0018] When the determination ratio is greater than or equal to the preset determination threshold, the pre-selected candidate region is the calibration target candidate region;

[0019] When the determination ratio is less than the preset determination threshold, the pre-selected candidate region is the background candidate region.

[0020] Optionally, the step of dividing the feature image into target candidate regions and background candidate regions through the region generation network model further includes:

[0021] A coordinate system is established based on the original image, and the position coordinates of the comparison image are obtained;

[0022] A coordinate system is established based on the image to be identified, and the position coordinates of all calibrated target candidate areas are obtained;

[0023] The difference between the position coordinates of all candidate calibrated target regions and the position coordinates of the comparison image is calculated to obtain the calibration difference of all candidate calibrated target regions;

[0024] All calibration differences are compared to obtain the calibration difference with the smallest value. The calibration target candidate region corresponding to the calibration difference with the smallest value is taken as the target candidate region, and the remaining calibration target candidate regions are taken as background candidate regions.

[0025] Optionally, the step of performing feature enhancement on the target candidate region to obtain the target candidate map includes:

[0026] Obtain each channel in the target candidate region, and calculate the maximum element value for each channel. The calculation method is as follows:

[0027] x max_pooling =max x i,j (i∈W,j∈H);

[0028] Where W represents the width of the target candidate region, H represents the length of the target candidate region, and x represents the width of the target candidate region. i,j This represents the element value at coordinate (i,j) of each channel in the target candidate region, x max_pooling This represents the maximum element value obtained from the target candidate region;

[0029] The mean of all calculated maximum element values ​​is obtained by averaging the maximum element values ​​of all channels. The averaging method is as follows:

[0030]

[0031] Where C represents the number of channels in the target candidate region, x max_pooling,i This represents the maximum element value obtained by the target candidate region in channel i. This represents the mean of the maximum element values ​​across all channels in the target candidate region.

[0032] The target candidate map is generated based on the mean of the maximum element values ​​of all channels.

[0033] Optionally, the step of performing feature reduction on the background candidate region to obtain a background candidate map includes:

[0034] Each channel in the background candidate region is obtained, and the average element value of each channel is calculated. The calculation method is as follows:

[0035]

[0036] Where W represents the width of the background candidate region, H represents the length of the background candidate region, and x represents the width of the background candidate region. i,j This represents the element value at coordinate (i,j) of each channel in the background candidate region, x avg_pooling This represents the average element value obtained from the background candidate region;

[0037] The mean of all calculated average element values ​​is obtained by averaging all the average element values ​​of all channels. The averaging method is as follows.

[0038]

[0039] Where C represents the number of channels in the background candidate region, x avg_pooling,i This represents the average element value obtained by the background candidate region in channel i. This represents the mean value of all channels in the background candidate region.

[0040] Background candidate images are generated based on the mean of the average element values ​​of all channels.

[0041] Optionally, the method further includes:

[0042] The background candidate image is removed, and the target candidate image is retained as the target for comparison with the comparison image.

[0043] Optionally, determining the qualified defect pre-selection box includes:

[0044] The defect pre-selection box is compared with the target candidate image and the background candidate image respectively;

[0045] When the similarity between the defect preselection box and the target candidate image is greater than the similarity between the defect preselection box and the background candidate image, the defect preselection box is determined to be a defect preselection box that meets the conditions.

[0046] When the similarity between the defect preselection box and the target candidate image is less than the similarity between the defect preselection box and the background candidate image, the defect preselection box is determined not to be a qualified defect preselection box.

[0047] In a second aspect, embodiments of this application provide a defect identification system for power equipment, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of any of the methods described in the first aspect above.

[0048] Thirdly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0049] Beneficial effects:

[0050] The present invention provides a method for identifying defects in power equipment. This method uses a region generation network model to divide the image to be identified, obtaining target candidate regions and background candidate regions. Feature enhancement and feature reduction are then performed on the target and background candidate regions respectively, resulting in target candidate images and background candidate images. These target and background candidate images can then be used to construct a defect identification model. This model is used to determine whether the image to be identified contains equipment defects. By constructing and training the defect identification model using the enhanced target candidate image and the reduced background candidate image, the trained model focuses on key information in the target candidate image while ignoring non-key information in the background candidate image. This eliminates interference from non-key information in defect identification, making equipment defect identification more accurate. Attached Figure Description

[0051] Figure 1 This is one of the flowcharts of the substation defect identification method according to a preferred embodiment of the present invention;

[0052] Figure 2 This is the second flowchart of a substation equipment defect identification method according to a preferred embodiment of the present invention. Detailed Implementation

[0053] The technical solution of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "connected" or "linked" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0055] Please see Figure 1-2 This application provides a method for identifying defects in power equipment, including:

[0056] Acquire the image to be identified, and extract the feature information from the image to be identified to generate a feature image;

[0057] The original image containing the equipment defect is acquired and labeled to obtain a comparison image containing the equipment defect. A region generation network model is then constructed based on the comparison image.

[0058] The feature image is divided into target candidate regions and background candidate regions using the region generation network model.

[0059] The target candidate region is enhanced to obtain a target candidate map, and the background candidate region is reduced to obtain a background candidate map.

[0060] A defect recognition model is constructed by using target candidate images and background candidate images. The image to be recognized is then input into the defect recognition model to obtain a series of defect pre-selection boxes, and the defect pre-selection boxes that meet the conditions are determined.

[0061] For the pre-selected defect boxes that meet the conditions, the defect value is calculated. When there is a pre-selected defect box in the image to be identified with a defect value greater than the preset threshold, the image to be identified is determined to have a defect. When there is no pre-selected defect box in the image to be identified with a defect value less than the preset threshold, the image to be identified is determined to have no defect.

[0062] In the above embodiments, the image to be identified is divided by a region generation network model to obtain target candidate regions and background candidate regions. Feature enhancement and feature reduction are performed on the target candidate regions and background candidate regions respectively to obtain target candidate maps and background candidate maps. The defect identification model can then be constructed using the target candidate maps and background candidate maps. The defect identification model is used to determine whether the image to be identified contains equipment defects. Constructing and training the defect identification model using the feature-enhanced target candidate maps and the feature-reduced background candidate maps enables the trained defect identification model to focus on the key information in the target candidate maps while ignoring the non-key information in the background candidate maps. This eliminates the interference of non-key information on defect identification and makes equipment defect identification more accurate.

[0063] The judgment process of the defect identification model is as follows:

[0064] 1) Input a new image to be recognized (WxHxC), pass it through Conv layers, and extract image features:

[0065] feature(x) = kernel k ·x (i,j,c) (k∈K,i∈W,j∈H,c∈C)

[0066] Where W represents the width of the input image, H represents the height of the input image, C represents the number of channels of the input image, kernel represents the convolution kernel, K represents the number of convolution kernels, and feature(x) represents the extracted image features.

[0067] 2) Input the extracted image features into the pre-trained defect recognition model to obtain a series of pre-selected boxes:

[0068] bbox = RPN(feature(x)) i (i∈K)

[0069] Where, feature(x) i Let represent the i-th extracted feature map, K represent the total number of feature maps, and bbox represent a series of pre-selected boxes.

[0070] 3) Classify and calculate the extracted pre-selected boxes to obtain the detection results:

[0071] label i =classifier(bbox) i (i∈N)

[0072] box i =bboxpredict(bbox) i (i∈N)

[0073] Where classifier represents the trained classification branch, bboxpredict represents the trained bounding box branch, and bbox... i The label represents the preselection boxes, where N represents the total number of preselection boxes. i This represents the category score of the i-th preselected box. i This represents the coordinate position of the i-th preselection box.

[0074] Based on the given category score threshold, determine whether the category score of the current series of preselected boxes meets the threshold. If it is greater than or equal to the given threshold, the preselected box is determined to have a defect. If it is less than the given threshold, the preselected box is determined to have no defect.

[0075] Optionally, the region generation network model based on the contrast image includes:

[0076] The comparison image is input into the region generation network to construct a region generation network model.

[0077] In the above embodiments, the region generation network model is trained by comparing the original image with a comparison image containing device defects, so that when the region generation network model is used to segment the image to be identified, it can more accurately identify the key regions containing device information in the image to be identified.

[0078] Optionally, the step of dividing the feature image into target candidate regions and background candidate regions through the region generation network model includes:

[0079] The feature image is divided using the region generation network model to obtain pre-selected candidate regions;

[0080] The pre-selected candidate region is overlapped with the comparison image to obtain the overlap area, and the area of ​​the pre-selected candidate region is added to the area of ​​the comparison image to obtain the judgment area;

[0081] The overlapping area is compared with the determined area to obtain a determination ratio, and the determination ratio is compared with a preset determination threshold.

[0082] When the determination ratio is greater than or equal to the preset determination threshold, the pre-selected candidate region is the calibration target candidate region;

[0083] When the determination ratio is less than the preset determination threshold, the pre-selected candidate region is the background candidate region.

[0084] In the above embodiments, when dividing the image, many pre-divided regions are obtained according to the division rules, and the divided regions are used as pre-selected candidate regions for screening. The overlapping area is obtained by overlaying the pre-selected candidate regions with the comparison image. The area of ​​the pre-selected candidate region is added to the area of ​​the comparison image to obtain the judgment area. The overlapping area is compared with the judgment area to obtain the judgment ratio. The preset judgment threshold can be determined according to the size of the comparison image marked during data annotation. When the judgment ratio is less than the preset judgment threshold, the pre-selected candidate region can be filtered out as the background candidate region. When the judgment ratio is greater than or equal to the preset judgment threshold, the pre-selected candidate region can be used as the target candidate region for the next step of target candidate region determination.

[0085] The above preliminary screening can reduce the number of calculations and obtain the target candidate region more intuitively and accurately.

[0086] Optionally, the step of dividing the feature image into target candidate regions and background candidate regions through the region generation network model further includes:

[0087] A coordinate system is established based on the original image, and the position coordinates of the comparison image are obtained;

[0088] A coordinate system is established based on the image to be identified, and the position coordinates of all calibrated target candidate areas are obtained;

[0089] The difference between the position coordinates of all candidate calibrated target regions and the position coordinates of the comparison image is calculated to obtain the calibration difference of all candidate calibrated target regions;

[0090] All calibration differences are compared to obtain the calibration difference with the smallest value. The calibration target candidate region corresponding to the calibration difference with the smallest value is taken as the target candidate region, and the remaining calibration target candidate regions are taken as background candidate regions.

[0091] In the above embodiments, by establishing a coordinate system, the calibration target candidate areas obtained after preliminary screening are transformed to obtain the coordinate values ​​of the calibration target candidate areas. The difference between the coordinate values ​​of the calibration target candidate areas and the coordinate values ​​of the comparison image is calculated to obtain the calibration difference value. The smaller the calibration difference value, the closer the calibration target candidate area is to the comparison image. Therefore, after comparing the calibration difference values, the smallest calibration difference value is obtained, which indicates that the calibration target candidate area corresponding to the smallest calibration difference value is a qualified target candidate area. The remaining calibration target candidate areas that exclude the qualified calibration target candidate area can be used as background candidate areas.

[0092] The method for establishing a coordinate system based on the original image is the same as the method for establishing a coordinate system based on the image to be recognized. Both methods set the left side of the image as the first side and the bottom side of the image as the second side, and take the intersection of the first and second sides as the origin of the coordinate axis. That is, the origin of the coordinate axis is the bottom point of the first side and the first point on the left side of the second side. The first side is taken as the vertical axis of the coordinate axis, i.e., the y-axis, and the second side is taken as the horizontal axis of the coordinate axis, i.e., the x-axis. This forms a coordinate system with the first side as the y-axis, the second side as the x-axis, and the intersection of the first and second sides as the origin.

[0093] Optionally, the step of performing feature enhancement on the target candidate region to obtain the target candidate map includes:

[0094] Obtain each channel in the target candidate region, and calculate the maximum element value for each channel. The calculation method is as follows:

[0095] x max_pooling =max x i,j (i∈W,j∈H);

[0096] Where W represents the width of the target candidate region, H represents the length of the target candidate region, and x represents the width of the target candidate region. i,j This represents the element value at coordinate (i,j) of each channel in the target candidate region, x max_pooling This represents the maximum element value obtained from the target candidate region;

[0097] The mean of all calculated maximum element values ​​is obtained by averaging the maximum element values ​​of all channels. The averaging method is as follows:

[0098]

[0099] Where C represents the number of channels in the target candidate region, x max_pooling,i This represents the maximum element value obtained by the target candidate region in channel i. This represents the mean of the maximum element values ​​across all channels in the target candidate region.

[0100] The target candidate map is generated based on the mean of the maximum element values ​​of all channels.

[0101] In the above embodiments, in order to obtain the image with enhanced features, after calculating the maximum element value of each channel and the mean of the maximum element values ​​of each channel, it is necessary to normalize the mean of the maximum element values ​​of each channel using a function to obtain function-normalized data. The function normalization process is as follows:

[0102]

[0103] in The sigmoid function represents the mean of the maximum element values ​​across all channels in the target candidate region. max_pooling (x) represents the normalized result of the mean of the maximum element values ​​of the channel.

[0104] Multiplying the normalized data by the target candidate boxes yields the target candidate map. The multiplication of the normalized data by the target candidate boxes is shown below:

[0105]

[0106] Where W represents the width of the target candidate region, H represents the length of the target candidate region, C represents the number of channels in the target candidate region, and x i,j,c This represents the element value at channel coordinates (i,j,c) in the target candidate region. This represents the result of multiplying the normalized data in the target candidate region by the element values ​​within the target candidate region.

[0107] Optionally, the step of performing feature reduction on the background candidate region to obtain a background candidate map includes:

[0108] Each channel in the background candidate region is obtained, and the average element value of each channel is calculated. The calculation method is as follows:

[0109]

[0110] Where W represents the width of the background candidate region, H represents the length of the background candidate region, and x represents the width of the background candidate region. i,j This represents the element value at coordinate (i,j) of each channel in the background candidate region, x avg_pooling This represents the average element value obtained from the background candidate region;

[0111] The mean of all calculated average element values ​​is obtained by averaging all the average element values ​​of all channels. The averaging method is as follows.

[0112]

[0113] Where C represents the number of channels in the background candidate region, xavg_pooling,i This represents the average element value obtained by the background candidate region in channel i. This represents the mean value of all channels in the background candidate region.

[0114] Background candidate images are generated based on the mean of the average element values ​​of all channels.

[0115] In the above embodiments, in order to obtain the image after feature reduction, after calculating the average element value of each channel and the mean of the average element values ​​of each channel, it is necessary to normalize the mean of the average element values ​​of each channel using a function to obtain function-normalized data. The function normalization process is as follows:

[0116]

[0117] in The sigmoid function represents the mean of the average element values ​​across all channels in the background candidate region. avg_pooling (x) represents the normalized result of the mean of the channel average element values;

[0118] Multiplying the normalized data by the background candidate boxes yields the background candidate image. The multiplication of the normalized data and the background candidate boxes is shown below:

[0119]

[0120] Where W represents the width of the background candidate region, H represents the length of the background candidate region, C represents the number of channels in the background candidate region, and x i,j,c This represents the element value at channel coordinates (i,j,c) in the background candidate region. This represents the result of multiplying the normalized data in the background candidate region by the element values ​​within the background candidate region.

[0121] Optionally, the method further includes:

[0122] The background candidate image is removed, and the target candidate image is retained as the target for comparison with the comparison image.

[0123] In the above embodiments, by eliminating background candidate images that do not contain key information, the recognition method can focus its attention on target candidate images that contain key information, thereby eliminating the influence of background candidate images on defect recognition and improving the accuracy of equipment defect recognition.

[0124] Optionally, determining the qualified defect pre-selection box includes:

[0125] The defect pre-selection box is compared with the target candidate image and the background candidate image respectively;

[0126] When the similarity between the defect preselection box and the target candidate image is greater than the similarity between the defect preselection box and the background candidate image, the defect preselection box is determined to be a defect preselection box that meets the conditions.

[0127] When the similarity between the defect preselection box and the target candidate image is less than the similarity between the defect preselection box and the background candidate image, the defect preselection box is determined not to be a qualified defect preselection box.

[0128] In the above embodiments, the size of the image to be judged is reduced by segmenting the image to be identified into defect pre-selection boxes. At the same time, the defect pre-selection boxes are compared with the target candidate image and the background candidate image to filter out defect pre-selection boxes with similar features to the background candidate image, thereby reducing the number of data images to be judged and improving the efficiency and accuracy of the calculation and judgment.

[0129] This application also provides a defect identification system for power equipment, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of any of the methods described in the above embodiments.

[0130] The above-described substation equipment defect identification system can implement various embodiments of the above-described substation equipment defect identification method and achieve the same beneficial effects, which will not be elaborated here.

[0131] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of any of the methods described in the above embodiments.

[0132] Various embodiments of the above-mentioned method for identifying defects in power equipment can be implemented and achieve the same beneficial effects, which will not be elaborated here.

[0133] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for identifying defects in power equipment, characterized in that, include: Acquire the image to be identified, and extract the feature information from the image to be identified to generate a feature image; The original image containing the equipment defect is acquired and labeled to obtain a comparison image containing the equipment defect. A region generation network model is then constructed based on the comparison image. The feature image is divided into target candidate regions and background candidate regions using the region generation network model. The target candidate region is enhanced to obtain a target candidate map, and the background candidate region is reduced to obtain a background candidate map. A defect recognition model is constructed by using target candidate images and background candidate images. The image to be recognized is then input into the defect recognition model to obtain a series of defect pre-selection boxes, and the defect pre-selection boxes that meet the conditions are determined. For the defect pre-selected boxes that meet the conditions, the defect value is calculated. When there is a defect pre-selected box in the image to be identified with a defect value greater than the preset threshold, the image to be identified is determined to have a defect. When there is no defect pre-selected box in the image to be identified with a defect value less than the preset threshold, the image to be identified is determined to have no defect. The step of dividing the feature image into target candidate regions and background candidate regions through the region generation network model includes: The feature image is divided using the region generation network model to obtain pre-selected candidate regions; The pre-selected candidate region is overlapped with the comparison image to obtain the overlap area, and the area of ​​the pre-selected candidate region is added to the area of ​​the comparison image to obtain the judgment area; The overlapping area is compared with the determined area to obtain a determination ratio, and the determination ratio is compared with a preset determination threshold. When the determination ratio is greater than or equal to the preset determination threshold, the pre-selected candidate region is the calibration target candidate region; When the determination ratio is less than the preset determination threshold, the pre-selected candidate region is the background candidate region; The step of dividing the feature image into target candidate regions and background candidate regions using the region generation network model further includes: A coordinate system is established based on the original image, and the position coordinates of the comparison image are obtained; A coordinate system is established based on the image to be identified, and the position coordinates of all calibrated target candidate areas are obtained; The difference between the position coordinates of all candidate calibrated target regions and the position coordinates of the comparison image is calculated to obtain the calibration difference of all candidate calibrated target regions; All calibration differences are compared to obtain the calibration difference with the smallest value. The calibration target candidate region corresponding to the calibration difference with the smallest value is taken as the target candidate region, and the remaining calibration target candidate regions are taken as background candidate regions.

2. The method for identifying defects in power equipment according to claim 1, characterized in that, The region generation network model based on contrast images includes: The comparison image is input into the region generation network to construct a region generation network model.

3. The method for identifying defects in power equipment according to claim 1, characterized in that, The step of performing feature enhancement on the target candidate region to obtain the target candidate map includes: Obtain each channel in the target candidate region, and calculate the maximum element value for each channel. The calculation method is as follows: ； Where W represents the width of the target candidate region, and H represents the length of the target candidate region. Represents the coordinates of each channel in the target candidate region. The element value, This represents the maximum element value obtained from the target candidate region; The mean of all calculated maximum element values ​​is obtained by averaging the maximum element values ​​of all channels. The averaging method is as follows: ； Where C represents the number of channels in the target candidate region. Indicates the target candidate region is in the channel The maximum element value obtained above, This represents the mean of the maximum element values ​​across all channels in the target candidate region. The target candidate map is generated based on the mean of the maximum element values ​​of all channels.

4. The method for identifying defects in power equipment according to claim 1, characterized in that, The step of performing feature reduction on the background candidate region to obtain a background candidate map includes: Each channel in the background candidate region is obtained, and the average element value of each channel is calculated. The calculation method is as follows: ； Where W represents the width of the background candidate region, and H represents the length of the background candidate region. Represents the coordinates of each channel in the background candidate region. The element value, This represents the average element value obtained from the background candidate region; The mean of all calculated average element values ​​is obtained by averaging all the average element values ​​of all channels. The averaging method is as follows. ； Where C represents the number of channels in the background candidate region. Indicates the background candidate area in the channel The average element value obtained above, This represents the mean value of all channels in the background candidate region. Background candidate images are generated based on the mean of the average element values ​​of all channels.

5. The method for identifying defects in power equipment according to claim 1, characterized in that, The method further includes: The background candidate image is removed, and the target candidate image is retained as the target for comparison with the comparison image.

6. The method for identifying defects in power equipment according to claim 1, characterized in that, The process of determining the qualified defect pre-selection box includes: The defect pre-selection box is compared with the target candidate image and the background candidate image respectively; When the similarity between the defect preselection box and the target candidate image is greater than the similarity between the defect preselection box and the background candidate image, the defect preselection box is determined to be a defect preselection box that meets the conditions. When the similarity between the defect preselection box and the target candidate image is less than the similarity between the defect preselection box and the background candidate image, the defect preselection box is determined not to be a qualified defect preselection box.

7. A defect identification system for power equipment, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1-6.

Images