A comprehensive evaluation method for strain clamps based on semantic segmentation and infrared analysis

Abstract

The present invention proposes a comprehensive evaluation method for strain clamps based on semantic segmentation and infrared analysis. The method uses visible light images and infrared temperature measurement images taken by an unmanned aerial vehicle, and utilizes a semantic segmentation model to extract the contour of the strain clamp in the visible light image; and generates a temperature matrix for the infrared image using an analytical algorithm; maps the contour of the strain clamp in the visible light image and the coordinates of each point inside it to the temperature matrix to obtain the temperature corresponding to each point of the strain clamp; then calculates the centroid of the strain clamp contour and the coordinates of the reference point; substitutes the calculated coordinates into the temperature matrix to calculate the highest temperature and the relative temperature difference; then uses a rust detection model to identify and segment the rust area of the strain clamp to obtain a mask image, and calculates the ratio of rust pixels; finally, comprehensively evaluates the operating state of the strain clamp according to the ratio of rust pixels and temperature defect information; the present invention effectively improves the recognition accuracy of the strain clamp and reduces the time-consuming of manual review.

Description

Technical Field

[0001] The present invention relates to the technical field of power line monitoring, and in particular to a comprehensive evaluation method for strain clamps based on semantic segmentation and infrared analysis. Background Art

[0002] A strain clamp is a device used to stabilize a transmission line and absorb the tension of the conductor, playing an important role in overhead transmission lines. However, in actual operation, there will be defects such as the heating of the drainage plate of the strain clamp, which is mainly caused by metal corrosion, metal aging, and sudden load changes. The traditional detection method is for front-line teams to regularly conduct inspections with infrared thermometers, which is costly. Now, drone inspections have been widely used in the power grid field, but different temperatures of the strain clamp show different characteristics in infrared thermography images. Even with the naked eye, it is difficult to identify its specific position in the infrared thermogram. Therefore, visible light images are introduced, and semantic segmentation technology is used to identify the strain clamp, extract its specific coordinates, map them to the temperature matrix calculated from the infrared thermogram, traverse the obtained temperature data to obtain the maximum temperature, minimum temperature, and average temperature. The image of the strain clamp is intercepted and identified with a rust detection model to obtain a mask image containing the rust result. After comprehensively analyzing the infrared temperature measurement results and the rust detection results, the status of the strain clamp is obtained. In terms of temperature: 3 - emergency defect; 2 - serious defect; 1 - general defect; 0 - no defect. In terms of rust: 2 - serious rust; 1 - general rust; 0 - no rust.

[0003] Chinese Patent CN115541050A discloses an overheating alarm device for strain clamps of high-voltage transmission lines, which includes a clamp temperature measurement device, a signal receiver, a temperature measurement monitoring host, and a background server. This device consists of many modules, showing that its structure is complex and the costs of deployment and maintenance are relatively high.

[0004] Chinese Patent CN218885168U discloses a temperature measurement device for line strain clamps, which includes a clamp drainage plate. The left and right upper parts of the clamp drainage plate are fixedly connected with fixing frames, and clamping mechanisms are inserted and movably connected to the opposite surfaces of the two fixing frames. Screws are threadedly connected to the middle parts of the upper ends of the two fixing frames. A temperature sensor is jointly clamped between the two clamping mechanisms. The lower end of the temperature sensor is fixedly connected with a mounting block. Two card slots are opened at the left and right ends of the mounting block. An installation slot is opened in the middle of the upper end of the clamp drainage plate, and movable slots are opened at the left and right upper parts of the clamp drainage plate. This device contains many components and modules, indicating that its deployment and maintenance are difficult. Summary of the Invention

[0005] The present invention proposes a comprehensive evaluation method for strain clamps based on semantic segmentation and infrared analysis, effectively improving the recognition accuracy of strain clamps and reducing the time-consuming of manual review.

[0006] The present invention adopts the following technical solutions.

[0007] A comprehensive evaluation method for strain clamps based on semantic segmentation and infrared analysis. The method uses visible light images and infrared temperature measurement images taken by a drone, and utilizes a semantic segmentation model to extract the contour of the strain clamp in the visible light image; and generates a temperature matrix for the infrared image using an analytical algorithm; maps the contour of the strain clamp in the visible light image and the coordinates of each point inside it to the temperature matrix to obtain the temperature corresponding to each point of the strain clamp; then calculates the centroid of the strain clamp contour and the coordinates of the reference point; substitutes the calculated coordinates into the temperature matrix to calculate the maximum temperature and the relative temperature difference; then uses a rust detection model to identify and segment the rust area of the strain clamp to obtain a mask image, and calculates the ratio of rust pixels; finally, comprehensively evaluates the operating state of the strain clamp based on the ratio of rust pixels and temperature defect information.

[0008] The evaluation method includes the following steps;

[0009] Step S1: Use a drone to take a visible light image and an infrared temperature measurement image of the strain clamp. The visible light image is a three-channel RGB image, and the infrared temperature measurement image is an original RAW format image. The infrared temperature measurement image is parsed by the DJI TSDK algorithm to obtain a temperature matrix;

[0010] Step S2: Input the visible light image into the strain clamp detection model PSPNet. The backbone network adopted by the model is MobileNet-v3-Large. After semantic segmentation, a mask image is obtained, the contour coordinates of the strain clamp inside the mask image are extracted, and the centroid coordinate M is calculated;

[0011] Step S3: Extract the reference point coordinates, calculate the maximum, minimum, and average temperatures according to the temperature matrix, and calculate the relative temperature difference;

[0012] Step S4: Extract the rust area, specifically, intercept a rectangular picture according to the coordinate contour of the strain clamp, and use DeepLabV3+ to extract the rust area inside;

[0013] Step S5: Comprehensively evaluate based on temperature and rust.

[0014] The size of the temperature matrix obtained in Step S1 is 640*512; the pixel size of the visible light image taken by the drone is 7680*6144. First, it is scaled to an image of 1280*1024 size, and then padded to an image of 1280*1280 size by boundary padding; the infrared temperature measurement image is an original RAW format image, and the pixel size of the image is 640*512;

[0015] In step S3, further call the DJI TSDK algorithm to calculate the infrared temperature measurement map, obtain the temperature matrix with a size of 640*512, and expand it to 1280*1024; then map the contour coordinates to the temperature matrix to obtain the temperature of the strain clamp, and obtain the temperatures of the three reference points in the same way. Select the point with the lowest temperature as the temperature reference point; and calculate the highest temperature, the lowest temperature, the average temperature and the relative temperature difference.

[0016] Step S2 includes the following steps;

[0017] Step S21: Input the visible light image into the strain clamp detection model PSPNet. The backbone network is MobileNet-v3-Large, and obtain the mask map through semantic segmentation;

[0018] Step S22: Extract the contour coordinates of the strain clamp from the mask map;

[0019] Step S23: Calculate the centroid coordinates according to the contour coordinates. The area formula A and the centroid coordinate formula M(C x ,C y ) are respectively:

[0020]

[0021]

[0022]

[0023] Step S4 includes the following steps,

[0024] Step S41: Calculate the coordinates of the three reference points according to the distance between the contour coordinates and the M point of the centroid coordinates, and then extract the coordinates of the three temperature reference points according to the contour coordinates and the center coordinates: Specifically:

[0025] First, calculate the coordinate P that is the farthest from the centroid coordinate in the contour coordinates 1 , and delete the coordinates in the contour coordinates whose included angle with P 1 M is less than 60°;

[0026] Second, calculate the coordinate P2 that is the farthest from M in the remaining coordinates, and delete the coordinates in the remaining coordinates whose included angle with P 2 M is less than 60°;

[0027] Third, calculate the coordinate P3 that is the farthest from M in the remaining coordinates; that is, P 1 , P 2 and P 3 are the coordinates of the three reference points;

[0028] Step S42: According to the temperature matrix and the profile coordinates, traverse the temperatures within the coordinates of the profile, and the highest, lowest, and average temperatures can be calculated;

[0029] Step S43: According to the temperature matrix, the highest temperature, and the temperature reference point coordinates, the relative temperature difference can be calculated. Among them, compare the temperatures of the three reference points, and select the lowest temperature as the reference point temperature τ 2 , according to the highest temperature τ 1 and the reference point temperature τ 2 , calculate the relative temperature difference δ:

[0030] δ = (τ 1 - τ 2 ) / τ1 * 100%.

[0031] Step S5 includes the following steps;

[0032] Step S51: According to the strain clamp profile coordinates, intercept the strain clamp picture;

[0033] Step S52: According to the strain clamp picture intercepted in Step S51, the corrosion detection model uses the DeepLabV3+ semantic segmentation model, and after semantic segmentation, a mask map reflecting the corrosion situation is obtained;

[0034] Step S53: According to the mask map in Step S52 and the profile coordinates obtained in Step S4, extract the pixels of the rust area, and calculate the ratio of the rust area to the pixels within the profile,

[0035] The calculation formula for the corrosion ratio corrosion_iou is corrosion_iou = corossion_area / area

[0036] where corossion_area is the rust pixel area and area is the pixel area of the strain clamp profile.

[0037] Step S6 includes the following steps;

[0038] Step S61: According to the relative temperature difference in Step 4, obtain the temperature situation of the strain clamp, and determine the defect type according to the temperature defect level determination table, including: emergency defect, serious defect, general defect, no defect.

[0039] Step S62: Synthesize the proportion of pixels with various corrosion degrees, compare the ratio of pixels with corrosion information to all pixel points within the strain clamp profile, and obtain the overall corrosion degree judgment of the strain clamp. The serious defect in Step S61 is that the rust pixel ratio exceeds 0.5; the general defect in Step S61 is that the rust pixel ratio exceeds 0.2; and the no defect in Step S61 is no rust;

[0040] Step S63: Based on Step S61 and Step S62, conduct a comprehensive evaluation and prompt the operation and maintenance personnel for defects with a score greater than 2 points.

[0041] The present invention provides a comprehensive evaluation method for strain clamps based on semantic segmentation and infrared analysis. For the visible light images and infrared thermography images taken by drones, the visible light images are directly given to the semantic segmentation model for recognition. The semantic segmentation model can classify each pixel and extract the contour coordinates of the entire strain clamp. The centroid of the strain clamp is extracted, and then temperature reference points are selected according to the centroid and contour coordinates. The infrared thermography image is used to generate a temperature matrix with the DJI TSDK algorithm. The contour coordinates of the strain clamp can be mapped into the temperature matrix to obtain the temperature of each pixel point of the strain clamp. Similarly, the temperature of the reference point can be obtained. The relative temperature difference can be calculated based on the highest temperature inside the strain clamp and the temperature of the reference point. Then, the strain clamp image is recognized using the corrosion detection model to obtain the corrosion detection result. By comprehensively analyzing the above two types of results, the temperature and corrosion status of the strain clamp can be obtained. The present invention effectively improves the recognition accuracy of the strain clamp and reduces the time-consuming of manual review.

[0042] The present invention has the following beneficial effects:

[0043] 1. This method is applicable to the visible light and infrared thermography images taken by drones, with low cost, and the drone can automatically inspect multiple devices at one time on the set flight route.

[0044] 2. This method uses the strain clamp detection model to classify each pixel and extract the contour of the strain clamp. After screenshot, the image of the strain clamp can be obtained.

[0045] 3. This method uses the DJI TSDK algorithm to calculate the temperature of the infrared thermography image and automatically extract the temperature of the reference point, providing a basis for relative temperature difference analysis.

[0046] 4. The corrosion detection model can be used to judge the corrosion condition and corrosion ratio of the strain clamp.

[0047] 5. The status of the strain clamp can be comprehensively determined according to the relative temperature difference and corrosion condition, which is more comprehensive, scientific and accurate than the traditional method of judging based on a single picture.

[0048] The present invention uses PSPNet to aggregate the context of different regions. PSPNet is a superior framework for pixel-level prediction, enabling the model to have the ability to understand global context information. Since the strain clamp often appears together with electrical equipment such as insulators, the electrical equipment that appears together at this time becomes the global information. PSPNet divides the obtained feature layer into regions of different sizes, and performs average pooling within each region respectively. By aggregating the context information of different regions, the ability to obtain global information is improved. The global information acquisition of the present invention can effectively produce high-quality results in scene analysis tasks. The contour of the strain clamp is extracted using a deep learning segmentation model, and then the segmentation of the rusty area is realized using the segmentation model. At the same time, a method for determining the degree of rust is designed; further, the temperature of the strain clamp is extracted by comprehensively analyzing the temperature of the infrared image; the degree of rust and the high-temperature area are comprehensively analyzed to determine the state of the clamp. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The present invention will be further described in detail below in conjunction with the drawings and specific embodiments:

[0050] FIG Figure 1 is a schematic flow chart of the present invention;

[0051] FIG Figure 2 is a schematic flow chart of the method for obtaining the mask map by segmenting the strain clamp;

[0052] FIG Figure 3 is a schematic flow chart of the method for extracting reference points;

[0053] FIG Figure 4 is a schematic diagram of the method for segmenting rust points on the strain clamp;

[0054] FIG Figure 5 is a schematic diagram of the comprehensive diagnosis of visible light and infrared images. SPECIFIC EMBODIMENTS

[0055] As shown in the figure, a comprehensive evaluation method for strain clamps based on semantic segmentation and infrared analysis. The method uses visible light images and infrared temperature measurement images taken by drones, and uses a semantic segmentation model to extract the contour of the strain clamp in the visible light image; and generates a temperature matrix for the infrared image using an analysis algorithm; maps the contour of the strain clamp in the visible light image and the coordinates of each point inside it to the temperature matrix to obtain the temperature corresponding to each point of the strain clamp; then calculates the centroid of the strain clamp contour and the reference point coordinates; substitutes the calculated coordinates into the temperature matrix to calculate the highest temperature and the relative temperature difference; then uses a rust detection model to identify and segment the rusty area of the strain clamp to obtain a mask map, and calculates the ratio of rust pixels; finally, based on the ratio of rust pixels and temperature defect information, comprehensively evaluate the operating state of the strain clamp.

[0056] The evaluation method includes the following steps;

[0057] Step S1: Use a drone to take visible light images and infrared thermometry images of the strain clamp. The visible light image is a three-channel RGB image, and the infrared thermometry image is an original RAW format image. The infrared thermometry image is parsed by the DJI TSDK algorithm to obtain a temperature matrix;

[0058] Step S2: Input the visible light image into the strain clamp detection model PSPNet. The backbone network used by the model is MobileNet-v3-Large. After semantic segmentation, a mask image is obtained. Extract the contour coordinates of the strain clamp in the mask image and calculate the centroid coordinate M;

[0059] Step S3: Extract the reference point coordinates, calculate the highest, lowest, and average temperatures according to the temperature matrix, and calculate the relative temperature difference;

[0060] Step S4: Extract the rust area. Specifically, according to the contour of the strain clamp coordinates, intercept a rectangular picture and use DeepLabV3+ to extract the rust area inside;

[0061] Step S5: Evaluate comprehensively based on temperature and rust.

[0062] The temperature matrix obtained in Step S1 has a size of 640*512; the pixel size of the visible light image taken by the drone is 7680*6144. First, scale it to an image with a size of 1280*1024, and then pad it to an image with a size of 1280*1280 through boundary padding; the infrared thermometry image is an original RAW format image, and the pixel size is 640*512;

[0063] In Step S3, further call the DJI TSDK algorithm to calculate the infrared thermometry map to obtain a temperature matrix with a size of 640*512, and expand it to 1280*1024; then map the contour coordinates to the temperature matrix to obtain the temperature of the strain clamp. Obtain the temperatures of the three reference points in the same way, and select the point with the lowest temperature as the temperature reference point; and calculate the highest temperature, lowest temperature, average temperature, and relative temperature difference.

[0064] Step S2 includes the following steps;

[0065] Step S21: Input the visible light image into the strain clamp detection model PSPNet. The backbone network is MobileNet-v3-Large, and a mask image is obtained through semantic segmentation;

[0066] Step S22: Extract the contour coordinates of the strain clamp in the mask image;

[0067] Step S23: Calculate the centroid coordinates based on the contour coordinates. The area formula A and the centroid coordinate formula M(C x ,C y ) involved are as follows:

[0068]

[0069]

[0070]

[0071] Step S4 includes the following steps.

[0072] Step S41: Calculate the coordinates of three reference points based on the distance between the contour coordinates and the centroid coordinate M point, and then extract the coordinates of three temperature reference points according to the contour coordinates and the center coordinates. Specifically:

[0073] First, calculate the coordinate P in the contour coordinates that is farthest from the centroid coordinate, 1 , and delete the coordinates in the contour coordinates whose included angle with P 1 M is less than 60°;

[0074] Second, calculate the coordinate P2 in the remaining coordinates that is farthest from M, and delete the coordinates in the remaining coordinates whose included angle with P 2 M is less than 60°;

[0075] Third, calculate the coordinate P3 in the remaining coordinates that is farthest from M; that is, P 1 , P 2 and P 3 are the coordinates of the three reference points;

[0076] Step S42: Traverse the temperatures within the coordinates of the contour according to the temperature matrix and the contour coordinates, and the highest, lowest, and average temperatures can be calculated;

[0077] Step S43: Calculate the relative temperature difference according to the temperature matrix, the highest temperature, and the temperature reference point coordinates. Among them, compare the temperatures of the three reference points, select the lowest temperature as the reference point temperature τ 2 , and calculate the relative temperature difference δ according to the highest temperature τ 1 and the reference point temperature τ 2 :

[0078] δ = (τ 1 -τ 2 ) / τ1 * 100%.

[0079] Step S5 includes the following steps;

[0080] Step S51: Intercept the picture of the strain clamp according to the contour coordinates of the strain clamp;

[0081] Step S52: According to the cut picture of the strain clamp in step S51, the corrosion detection model uses the DeepLabV3+ semantic segmentation model. After semantic segmentation, a mask graph reflecting the corrosion situation is obtained;

[0082] Step S53: According to the mask graph in step S52 and the contour coordinates obtained in step S4, extract the pixels of the corroded area and calculate the ratio of the pixels in the corroded area to the pixels within the contour.

[0083] The calculation formula for the corrosion ratio corrosion_iou is corrosion_iou = corossion_area / area

[0084] Where corossion_area is the pixel area of the corrosion, and area is the pixel area of the strain clamp contour.

[0085] Step S6 includes the following steps;

[0086] Step S61: Based on the relative temperature difference in step 4, obtain the temperature condition of the strain clamp, and determine the defect type according to the temperature defect level determination table, including: emergency defect, serious defect, general defect, no defect.

[0087] Step S62: Synthesize the proportion of pixels with various corrosion degrees, compare the ratio of pixels with corrosion information to all pixel points within the strain clamp contour, and obtain the overall corrosion degree judgment of the strain clamp. A serious defect in step S61 is that the proportion of corrosion pixels exceeds 0.5; a general defect in step S61 is that the proportion of corrosion pixels exceeds 0.2; no defect in step S61 means no corrosion;

[0088] Step S63: According to steps S61 and S62, conduct a comprehensive evaluation using Tables 1, 2, and 3 below, and prompt the maintenance personnel for defects with a score greater than 2 points.

[0089]

[0090] Table 1 Temperature Defect Judgment Table

[0091]

[0092] Table 2 Corrosion Grade Judgment Table

[0093]

[0094] Table 3 Comprehensive Evaluation Judgment Table.

Claims

1. A comprehensive evaluation method for strain clamps based on semantic segmentation and infrared analysis, characterized in that: The method uses visible light images and infrared temperature measurement images taken by a drone. The semantic segmentation model is used to extract the contour of the strain clamp in the visible light image; and an analytical algorithm is used to generate a temperature matrix for the infrared image; the contour of the strain clamp in the visible light image and the coordinates of each point inside it are mapped into the temperature matrix to obtain the temperature corresponding to each point of the strain clamp; then the centroid of the strain clamp contour and the reference point coordinates are calculated; the calculated coordinates are substituted into the temperature matrix to calculate the highest temperature and the relative temperature difference; then the rust detection model is used for identification, the rust area of the strain clamp is segmented to obtain a mask image, and the ratio of rust pixels is calculated; finally, based on the ratio of rust pixels and the temperature defect information, the operating state of the strain clamp is comprehensively evaluated; The evaluation method includes the following steps; Step S1: Use a drone to take visible light images and infrared temperature measurement images of the strain clamp. The visible light image is a three-channel RGB image, and the infrared temperature measurement image is an original RAW format image. The infrared temperature measurement image is analyzed by the DJI TSDK algorithm to obtain a temperature matrix; Step S2: Input the visible light image into the strain clamp detection model PSPNet. The backbone network used by the model is MobileNet-v3-Large. After semantic segmentation, a mask image is obtained, the contour coordinates of the strain clamp inside the mask image are extracted, and the centroid coordinate M is calculated; Step S3: Extract the reference point coordinates, calculate the highest, lowest, and average temperatures according to the temperature matrix, and calculate the relative temperature difference; Step S4: Extract the rust area, specifically, according to the coordinate contour of the strain clamp, intercept a rectangular picture, and use DeepLabV3+ to extract the rust area inside; Step S5: Conduct a comprehensive evaluation based on temperature and rust; Step S5 includes the following steps; Step S61: According to the relative temperature difference in Step 4, obtain the temperature condition of the strain clamp, and determine the defect type according to the temperature defect grade determination table, including: emergency defect, serious defect, general defect, no defect; Step S62: Comprehensively consider the proportion of pixels with each rust degree, compare the ratio of pixels with rust information to all pixel points inside the contour of the strain clamp, and obtain the overall rust degree judgment of the strain clamp. The serious defect in Step S61 is that the ratio of rust pixels exceeds 0.5; the general defect in Step S61 is that the ratio of rust pixels exceeds 0.2; the no defect in Step S61 is no rust; Step S63: According to Step S61 and Step S62, conduct a comprehensive evaluation, and prompt the maintenance personnel for defects with a score greater than 2 points.

2. A comprehensive evaluation method for strain clamps based on semantic segmentation and infrared analysis according to claim 1, characterized in that: Step S1 obtains a temperature matrix with a size of 640*512; the pixel size of the visible light image captured by the drone is 7680*6144. First, it is scaled into an image with a size of 1280*1024, and then padded to an image with a size of 1280*1280 through boundary padding; the infrared thermometry image is the original RAW format image, and the image pixel size is 640*512. In step S3, the DJI TSDK algorithm is called to calculate the infrared thermometry map to obtain a temperature matrix with a size of 640*512, which is expanded to 1280*1024; then the contour coordinates are mapped into the temperature matrix to obtain the temperature of the strain clamp. The temperatures of the three reference points are obtained in the same way, and the point with the lowest temperature is selected as the temperature reference point; and the highest temperature, the lowest temperature, the average temperature and the relative temperature difference are calculated.

3. A comprehensive evaluation method for strain clamps based on semantic segmentation and infrared analysis according to claim 1, characterized in that: Step S2 includes the following steps; Step S21: The visible light image is input into the strain clamp detection model PSPNet, and the backbone network is MobileNet-v3-Large. After semantic segmentation, a mask image is obtained; Step S22: Extract the contour coordinates of the strain clamp in the mask image; Step S23: Calculate the centroid coordinates based on the contour coordinates. The area formula A and the centroid coordinate formula M(C x , C y ) involved are respectively as follows:

4. A comprehensive evaluation method for strain clamps based on semantic segmentation and infrared analysis according to claim 1, characterized in that: Step S3 includes the following steps, Step S41: Calculate the coordinates of the three reference points according to the distance between the contour coordinates and the center of gravity coordinate M point, and then extract the coordinates of the three temperature reference points according to the contour coordinates and the center coordinates: specifically: First, calculate the coordinate P in the contour coordinates that is farthest from the centroid coordinate 1 , and delete the coordinates in the contour coordinates that form an angle less than 60° with P 1 and M; Second, calculate the coordinate P2 that is farthest from M among the remaining coordinates, and delete the coordinates in the remaining coordinates whose included angle with P 2 M is less than 60°; Third, calculate the coordinate P3 that is farthest from M among the remaining coordinates; that is, P 1 , P 2 and P 3 are the coordinates of three reference points; Step S42: According to the temperature matrix and the contour coordinates, traverse the temperatures within the contour coordinates to calculate the highest, lowest and average temperatures; Step S43: Calculate the relative temperature difference according to the temperature matrix, the highest temperature, and the temperature reference point coordinates. Among them, compare the temperatures of the three reference points, and select the lowest temperature as the reference point temperature τ 2， According to the highest temperature τ 1 and the reference point temperature τ 2 , calculate the relative temperature difference δ: δ = (τ 1 - τ 2 ) / τ1 * 100%.

5. A comprehensive evaluation method for strain clamps based on semantic segmentation and infrared analysis according to claim 1, characterized in that: Step S4 includes the following steps; Step S51: Intercept the strain clamp picture according to the strain clamp contour coordinates; Step S52: According to the strain clamp picture intercepted in step S51, the corrosion detection model uses the DeepLabV3+ semantic segmentation model, and a mask image reflecting the corrosion situation is obtained after semantic segmentation; Step S53: According to the mask image in step S52 and the contour coordinates obtained in step S4, extract the pixels in the corrosion area, and calculate the ratio of the corrosion area to the pixels within the contour, The corrosion occupancy ratio corrosion_iou calculation formula is corrosion_iou = corossion_area / area where corossion_area is the area of the corrosion pixels and area is the area of the pixels of the strain clamp contour.

Images