Power equipment surface defect detection method and system based on machine learning

Abstract

The invention discloses a method and system for detecting surface defects of power equipment based on machine learning, and belongs to the technical field of defect detection.The method specifically comprises the steps that a target area is set, and an infrared monitoring video and an optical monitoring video of a power transmission line are obtained in real time through an unmanned aerial vehicle; the method comprises the following steps of: framing an infrared monitoring video according to a preset frame length, selecting any infrared image and inputting the selected infrared image into a preset equipment identification model to obtain an equipment image, generating a positioning frame of any power transmission line, acquiring an infrared temperature and a positioning frame width of any power transmission line, constructing a point location map by taking the positioning frame width as a horizontal coordinate and the infrared temperature as a vertical coordinate, and identifying the equipment image according to the point location map. And fitting the point map to obtain a fitting function, correcting the infrared temperature of any power transmission line, determining a target infrared image, extracting an optical image corresponding to the target infrared image, and substituting the optical image into a preset defect identification model to obtain an equipment defect identification result of the power transmission line.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection, and in particular to a method and system for detecting surface defects of power equipment based on machine learning. Background Art

[0002] Surface defect detection of power equipment is an important part of power grid safety production and a key step in the automatic inspection of intelligent patrol systems. With the help of surface defect detection technology, the quality and efficiency of intelligent patrol systems can be effectively improved. Common processing methods include: 1) Traditional algorithms; Traditional surface defect detection algorithms obtain images that are easy to detect through image preprocessing, and then use statistical machine learning methods to extract image features to achieve the goal of defect detection. 2) Deep learning; Deep learning algorithms use deep convolutional neural networks (DCNNs) combined with SSD, Yolo and other network methods to build a coarse-to-fine cascade detection network, including equipment positioning, defect detection and classification.

[0003] In recent years, drone inspection technology has gradually been applied to the inspection of power transmission and distribution equipment. The optical monitoring video and infrared monitoring video of the inspection area are collected by drones in real time, so as to realize remote monitoring of power transmission and distribution equipment. However, as the shooting distance increases, the area covered by each pixel in the infrared image will also increase accordingly, resulting in a decrease in the proportion of the measured object in the field of view, which in turn reduces the quality of the infrared radiation output signal and reduces the temperature measurement accuracy. In addition, the longer the distance, the greater the impact of the air on the infrared radiation, resulting in an increase in the infrared temperature measurement error, which may lead to misjudgment. As a result, a large amount of computing power is often required to check the optical monitoring video, and the actual efficiency is low. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for detecting surface defects of power equipment based on machine learning to solve the following technical problems:

[0005] In recent years, drone inspection technology has gradually been applied to the inspection of power transmission and distribution equipment. The optical monitoring video and infrared monitoring video of the inspection area are collected by drones in real time, so as to realize remote monitoring of power transmission and distribution equipment. However, as the shooting distance increases, the area covered by each pixel in the infrared image will also increase accordingly, resulting in a decrease in the proportion of the measured object in the field of view, which in turn reduces the quality of the infrared radiation output signal and reduces the temperature measurement accuracy. In addition, the farther the distance, the greater the impact of the air on the infrared radiation, resulting in an increase in the infrared temperature measurement error, which may lead to misjudgment. As a result, a large amount of computing power is often required to check the optical monitoring video, which is inefficient.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A method for detecting surface defects of power equipment based on machine learning, characterized in that it comprises the following steps:

[0008] S1, set the target area, obtain the power transmission lines in the target area and mark them as the inspection route of the drone, and obtain the infrared monitoring video and optical monitoring video of the power transmission lines in real time through the drone;

[0009] S2, divide the infrared monitoring video into frames according to a preset frame length to obtain an infrared image sequence y1, y2, ..., yN, select any infrared image and input it into a preset device recognition model to obtain a device image, generate a positioning frame of any transmission line from the device image, obtain the infrared temperature and positioning frame width of any transmission line, construct a point map with the positioning frame width as the horizontal coordinate and the infrared temperature as the vertical coordinate, fit the point map to obtain a fitting function T(D), and the fitting function T(D) is used to fit the relationship between the infrared temperature and the positioning frame width;

[0010] S3, correct the infrared temperature of any transmission line according to the fitting function T(D), determine the target infrared image according to the corrected infrared temperature, extract the optical image corresponding to the target infrared image and mark it as the target optical image, substitute the target optical image into the preset defect recognition model, and obtain the equipment defect recognition result of the transmission line.

[0011] As a further solution of the present invention: In S2, the specific process of generating a positioning frame of any transmission line is as follows:

[0012] The device image is gray-processed to obtain a device gray-scale image, the device gray-scale image is edge-detected to obtain a number of edge regions, any edge region is selected and edge extraction is performed to obtain edge features, straight line features are extracted from the edge features and the straight line features are simplified by using the Douglas–ρeucker algorithm; any straight line segment is selected and the length and direction of the straight line segment are calculated to obtain the lengths and directions of all straight line segments, all straight line segments in a preset direction interval [0, 180] are obtained, the preset direction interval is divided into a number of sub-direction intervals according to a preset step size, and the straight line eigenvalue ρ and the eigenvalue ρ in any sub-direction interval are calculated in turn. Where m is the number of straight lines in sub-direction interval j, n is the number of all sub-direction intervals, and qi is the length of any straight line segment;

[0013] If there is any sub-direction interval whose straight line eigenvalue ρ is greater than a preset threshold, the sub-direction interval is marked as a target direction interval; the direction difference between the target direction intervals is calculated, and if there is a direction difference of 90 degrees, the edge area is marked as a background area; if there is no direction difference of 90 degrees between the target direction intervals, the edge area is marked as a transmission line area, a minimum circumscribed rectangle is circumscribed to the transmission line area, and the pixel rectangle is marked as the positioning frame of the transmission line.

[0014] As a further solution of the present invention: in S2, the infrared temperature of the transmission line is the highest temperature value in the transmission line area.

[0015] As a further solution of the present invention: S2 further includes screening the infrared image, and the specific screening process is:

[0016] Obtain the infrared temperature of all transmission lines in any infrared image, calculate the range C and temperature dispersion K corresponding to the infrared image, use the temperature dispersion as the horizontal coordinate and the range as the vertical coordinate, generate feature points corresponding to all infrared images, select any feature point as the center, set the clustering control radius R, obtain the feature point density ρ within the control radius R of any feature point, calculate the feature point density mean Avgρ, if there is any feature point with a feature point density ρ greater than Avgρ within the radius R, then use the feature point as the core point and generate a category cluster, generate several category clusters, obtain the number of feature points in any category cluster, select the category cluster with the largest number of feature points and mark it as the target category cluster, use the infrared image corresponding to the target category cluster as the target infrared image, obtain the infrared temperature of the transmission line and the positioning frame width in any target infrared image for function fitting;

[0017] The specific calculation formula is:

[0018] C = MAXT - MINT;

[0019]

[0020] Where n is the total number of transmission lines in the infrared image, MAXT is the maximum value of the transmission line temperature in the infrared image, and MINT is the minimum value of the transmission line temperature in the infrared image.

[0021] As a further solution of the present invention: the specific process of setting the cluster control radius R is:

[0022] Taking any feature point as the center, calculate the Euclidean distance d between the feature point and any other feature point, sum each Euclidean distance to get u, and get the control radius R based on the value u. The calculation formula is as follows:

[0023]

[0024] Among them, u is the sum of the Euclidean distance data values ​​of all feature points, and d is the Euclidean distance between any feature points.

[0025] As a further solution of the present invention: if there is a non-core point in any category cluster and the density ρ within the control radius R is also greater than Avgρ, the category cluster generated by the non-core point is merged with the original category cluster to generate several category clusters.

[0026] As a further solution of the present invention: in S4, the specific determination process of the target infrared image is:

[0027] S11, determine the standard positioning frame width D 标 , obtain the positioning frame width Da and infrared temperature Ta of any transmission line, and calculate D 标 The difference △H between the infrared temperature Ta of the transmission line and Da is substituted into the fitting function T(D) to calculate the fitting temperature T', and the corrected temperature Tb is obtained by summing the current infrared temperature Ta of the transmission line and the fitting temperature T'.

[0028] S12, repeat the above step S11 to obtain the corrected temperature Tb of all transmission lines in any infrared image, and calculate the corrected temperature difference between any transmission line and all other transmission lines in turn. If any corrected temperature difference is greater than or equal to a preset threshold, the infrared image is calibrated as a target infrared image.

[0029] A detection system for surface defects of power equipment based on machine learning, characterized by comprising:

[0030] The data acquisition module is used to set the target area, acquire the power transmission lines in the target area and mark them as the inspection route of the drone, and acquire the infrared monitoring video and optical monitoring video of the power transmission lines in real time through the drone;

[0031] A data analysis module is used to perform frame processing on the infrared monitoring video according to a preset frame length to obtain an infrared image sequence y1, y2, ..., yN, select any infrared image and input it into a preset device recognition model to obtain a device image, generate a positioning frame of any transmission line from the device image, obtain the infrared temperature and positioning frame width of any transmission line, construct a point map with the positioning frame width as the horizontal coordinate and the infrared temperature as the vertical coordinate, fit the point map to obtain a fitting function T(D), and the fitting function T(D) is used to fit the relationship between the infrared temperature and the positioning frame width;

[0032] The result generation module is used to correct the infrared temperature of any transmission line according to the fitting function T(D), determine the target infrared image according to the corrected infrared temperature, extract the optical image corresponding to the target infrared image and mark it as the target optical image, substitute the target optical image into the preset defect recognition model, and obtain the equipment defect recognition result of the transmission line.

[0033] Beneficial effects of the present invention:

[0034] The present invention formulates an inspection route for a drone, collects infrared monitoring videos and optical monitoring videos of the inspection area in real time, and performs temperature correction through a fitting function of the relationship between the positioning frame width and the infrared temperature in the infrared image. It can be understood that the positioning frame width represents the shooting distance between the transmission line and the drone. The closer the shooting distance, the larger the positioning frame width, and the farther the shooting distance, the smaller the positioning frame width. At the same time, the transmission line belongs to a three-phase device, so the surface temperature difference method can be used to compare and analyze the temperature of corresponding parts of similar devices. When the hot spot feature of the infrared image of a certain phase device is obviously different from that of the other two phases, it can be determined that the phase is suspected to be faulty. However, the shooting distance of each transmission line is not the same during the actual shooting process, so there is an infrared temperature error. The present invention uses the positioning frame width as the horizontal coordinate and the infrared temperature as the vertical coordinate to construct a point map and fit the point map to obtain a fitting function. The infrared temperature is corrected according to the fitting function, the influence of environmental factors and shooting distance on the infrared temperature is reduced, and the accuracy of infrared temperature data is improved. The target infrared image is determined according to the corrected infrared temperature to perform preliminary defect screening, and the optical image is used for detail confirmation and defect confirmation. The combination of the two can more comprehensively assess the status of transmission line equipment and avoid misjudgment caused by a single image type. It first identifies possible risks of high temperature, fire and electric sparks through infrared monitoring, and then further accurately identifies them through optical images, which not only improves inspection efficiency but also saves computing power. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The present invention will be further described below in conjunction with the accompanying drawings.

[0036] Figure 1 It is a flow chart of a method for detecting surface defects of power equipment based on machine learning of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] See also Figure 1 As shown, the present invention is a method for detecting surface defects of power equipment based on machine learning, comprising the following steps:

[0039] S1, set the target area, obtain the power transmission lines in the target area and mark them as the inspection route of the drone, and obtain the infrared monitoring video and optical monitoring video of the power transmission lines in real time through the drone;

[0040] S2, divide the infrared monitoring video into frames according to a preset frame length to obtain an infrared image sequence y1, y2, ..., yN, select any infrared image and input it into a preset device recognition model to obtain a device image, generate a positioning frame of any transmission line from the device image, obtain the infrared temperature and positioning frame width of any transmission line, construct a point map with the positioning frame width as the horizontal coordinate and the infrared temperature as the vertical coordinate, fit the point map to obtain a fitting function T(D), and the fitting function T(D) is used to fit the relationship between the infrared temperature and the positioning frame width;

[0041] S3, correct the infrared temperature of any transmission line according to the fitting function T(D), determine the target infrared image according to the corrected infrared temperature, extract the optical image corresponding to the target infrared image and mark it as the target optical image, substitute the target optical image into the preset defect recognition model, and obtain the equipment defect recognition result of the transmission line.

[0042] The present invention formulates an inspection route for a drone, collects infrared monitoring videos and optical monitoring videos of the inspection area in real time, and performs temperature correction through a fitting function of the relationship between the positioning frame width and the infrared temperature in the infrared image. It can be understood that the positioning frame width represents the shooting distance between the transmission line and the drone. The closer the shooting distance, the larger the positioning frame width, and the farther the shooting distance, the smaller the positioning frame width. At the same time, the transmission line belongs to a three-phase device, so the surface temperature difference method can be used to compare and analyze the temperature of corresponding parts of similar devices. When the hot spot feature of the infrared image of a certain phase device is obviously different from that of the other two phases, it can be determined that the phase is suspected to be faulty. However, the shooting distance of each transmission line is not the same during the actual shooting process, so there is an infrared temperature error. The present invention uses the positioning frame width as the horizontal coordinate and the infrared temperature as the vertical coordinate to construct a point map and fit the point map to obtain a fitting function. The infrared temperature is corrected according to the fitting function, the influence of environmental factors and shooting distance on the infrared temperature is reduced, and the accuracy of infrared temperature data is improved. The target infrared image is determined according to the corrected infrared temperature to perform preliminary defect screening, and the optical image is used for detail confirmation and defect confirmation. The combination of the two can more comprehensively assess the status of transmission line equipment and avoid misjudgment caused by a single image type. It first identifies possible risks of high temperature, fire and electric sparks through infrared monitoring, and then further accurately identifies them through optical images, which not only improves inspection efficiency but also saves computing power.

[0043] It is worth noting that scratches, wear or broken strands on the surface of the wires in overhead transmission lines will destroy their integrity, increase resistance and cause local overheating. In addition, exposed wires are prone to oxidation reactions, which further aggravate temperature anomalies. Therefore, for transmission lines with surface defects, there will be local overheating. At the same time, overhead lines are three-phase equipment, that is, the voltage of each transmission line is the same, so the actual temperature of each transmission line is roughly the same, but the transmission lines in actual overhead transmission lines are not on the same horizontal plane, and the drone is inspected and photographed at a fixed height. Therefore, it is affected by the shooting distance and the actual environment, and there will be infrared temperature measurement errors, resulting in misjudgment. Therefore, the present invention determines all transmission line data during the inspection process, and uses the positioning frame width as the horizontal coordinate and the infrared temperature as the vertical coordinate to construct a point map, fit the point map, obtain the fitting function, and obtain the relationship between the infrared temperature and the positioning frame width.

[0044] According to the fitting function, the corrected temperature Tb of all transmission lines in any infrared image is obtained, and the corrected temperature difference between any transmission line and all other transmission lines is calculated in turn. If any corrected temperature difference is greater than or equal to a preset threshold, the infrared image is calibrated as a target infrared image. It can be understood that the overhead line is a three-phase device, that is, the voltage of each transmission line is the same, so the actual temperature of each transmission line is roughly the same. When there is a temperature difference greater than or equal to the preset threshold, it indicates that the corresponding transmission line in the infrared image is abnormal. Therefore, the infrared image is calibrated as a target infrared image, and then the optical image corresponding to the target infrared image is extracted, and the target optical image is substituted into the preset defect recognition model to obtain the equipment defect recognition result of the transmission line. The preset defect recognition model is a convolutional neural network (CNN) model based on deep learning technology. The model can automatically extract features from infrared images and complete classification and recognition. It will not be repeated here, thereby greatly reducing the recognition computing power and improving the recognition efficiency.

[0045] In a preferred embodiment of the present invention, in said S2, the specific process of generating a positioning frame of any transmission line is:

[0046] The device image is gray-processed to obtain a device gray-scale image, the device gray-scale image is edge-detected to obtain a number of edge regions, any edge region is selected and edge extraction is performed to obtain edge features, straight line features are extracted from the edge features and the straight line features are simplified by using the Douglas–ρeucker algorithm; any straight line segment is selected and the length and direction of the straight line segment are calculated to obtain the lengths and directions of all straight line segments, all straight line segments whose directions are within a preset direction interval [0, 180] are obtained, the preset direction interval is divided into a number of sub-direction intervals according to a preset step size, and the straight line eigenvalue ρ, the eigenvalue ρ in any sub-direction interval are calculated in turn. Where m is the number of straight lines in sub-direction interval j, n is the number of all sub-direction intervals, and qi is the length of any straight line segment;

[0047] If there is any sub-direction interval whose straight line eigenvalue ρ is greater than a preset threshold, the sub-direction interval is marked as a target direction interval; the direction difference between the target direction intervals is calculated, and if there is a direction difference of 90 degrees, the edge area is marked as a background area; if there is no direction difference of 90 degrees between the target direction intervals, the edge area is marked as a transmission line area, a minimum circumscribed rectangle is circumscribed to the transmission line area, and the pixel rectangle is marked as the positioning frame of the transmission line.

[0048] In another preferred embodiment of the present invention, in S2, the infrared temperature of the transmission line is the highest temperature value in the transmission line area.

[0049] In another preferred embodiment of the present invention, the step S2 further includes screening the infrared image, and the specific process of screening is:

[0050] Obtain the infrared temperature of all transmission lines in any infrared image, calculate the range C and temperature dispersion K corresponding to the infrared image, use the temperature dispersion as the horizontal coordinate and the range as the vertical coordinate, generate feature points corresponding to all infrared images, select any feature point as the center, set the clustering control radius R, obtain the feature point density ρ within the control radius R of any feature point, calculate the feature point density mean Avgρ, if there is any feature point with a feature point density ρ greater than Avgρ within the radius R, then use the feature point as the core point and generate a category cluster, generate several category clusters, obtain the number of feature points in any category cluster, select the category cluster with the largest number of feature points and mark it as the target category cluster, use the infrared image corresponding to the target category cluster as the target infrared image, obtain the infrared temperature of the transmission line and the positioning frame width in any target infrared image for function fitting;

[0051] The specific calculation formula is:

[0052] C = MAXT - MINT;

[0053]

[0054] Where n is the total number of transmission lines in the infrared image, MAXT is the maximum value of the transmission line temperature in the infrared image, and MINT is the minimum value of the transmission line temperature in the infrared image.

[0055] In another preferred embodiment of the present invention, the specific process of setting the cluster control radius R is:

[0056] Taking any feature point as the center, calculate the Euclidean distance d between the feature point and any other feature point, sum each Euclidean distance to get u, and get the control radius R based on the value u. The calculation formula is as follows:

[0057]

[0058] Among them, u is the sum of the Euclidean distance data values ​​of all feature points, and d is the Euclidean distance between any feature points.

[0059] In another preferred embodiment of the present invention, if there is a non-core point in any category cluster and the density ρ within the control radius R is also greater than Avgρ, the category cluster generated by the non-core point is merged with the original category cluster to generate several category clusters.

[0060] In another preferred embodiment of the present invention, in S4, the specific process of determining the target infrared image is:

[0061] S11, determine the standard positioning frame width D 标 , obtain the positioning frame width Da and infrared temperature Ta of any transmission line, and calculate D 标 The difference △H between the infrared temperature Ta of the transmission line and Da is substituted into the fitting function T(D) to calculate the fitting temperature T', and the corrected temperature Tb is obtained by summing the current infrared temperature Ta of the transmission line and the fitting temperature T'.

[0062] S12, repeat the above step S11 to obtain the corrected temperature Tb of all transmission lines in any infrared image, and calculate the corrected temperature difference between any transmission line and all other transmission lines in turn. If any corrected temperature difference is greater than or equal to a preset threshold, the infrared image is calibrated as a target infrared image.

[0063] A detection system for surface defects of power equipment based on machine learning, comprising:

[0064] The data acquisition module is used to set the target area, acquire the power transmission lines in the target area and mark them as the inspection route of the drone, and acquire the infrared monitoring video and optical monitoring video of the power transmission lines in real time through the drone;

[0065] A data analysis module is used to perform frame processing on the infrared monitoring video according to a preset frame length to obtain an infrared image sequence y1, y2, ..., yN, select any infrared image and input it into a preset device recognition model to obtain a device image, generate a positioning frame of any transmission line from the device image, obtain the infrared temperature and positioning frame width of any transmission line, construct a point map with the positioning frame width as the horizontal coordinate and the infrared temperature as the vertical coordinate, fit the point map to obtain a fitting function T(D), and the fitting function T(D) is used to fit the relationship between the infrared temperature and the positioning frame width;

[0066] The result generation module is used to correct the infrared temperature of any transmission line according to the fitting function T(D), determine the target infrared image according to the corrected infrared temperature, extract the optical image corresponding to the target infrared image and mark it as the target optical image, substitute the target optical image into the preset defect recognition model, and obtain the equipment defect recognition result of the transmission line.

[0067] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for detecting surface defects of power equipment based on machine learning, characterized in that: The following steps are involved: S1, set the target area, obtain the power transmission lines in the target area and mark them as the inspection route of the drone, and obtain the infrared monitoring video and optical monitoring video of the power transmission lines in real time through the drone; S2, divide the infrared monitoring video into frames according to a preset frame length to obtain an infrared image sequence y1, y2, ..., yN, select any infrared image and input it into a preset device recognition model to obtain a device image, generate a positioning frame of any transmission line from the device image, obtain the infrared temperature and positioning frame width of any transmission line, construct a point map with the positioning frame width as the horizontal coordinate and the infrared temperature as the vertical coordinate, fit the point map to obtain a fitting function T(D), and the fitting function T(D) is used to fit the relationship between the infrared temperature and the positioning frame width; S3, correct the infrared temperature of any transmission line according to the fitting function T(D), determine the target infrared image according to the corrected infrared temperature, extract the optical image corresponding to the target infrared image and mark it as the target optical image, substitute the target optical image into the preset defect recognition model, and obtain the equipment defect recognition result of the transmission line.

2. The method for detecting surface defects of power equipment based on machine learning according to claim 1, characterized in that: In S2, the specific process of generating a positioning frame of any transmission line is as follows: The device image is gray-processed to obtain a device gray-scale image, the device gray-scale image is edge-detected to obtain a number of edge regions, any edge region is selected and edge extraction is performed to obtain edge features, straight line features are extracted from the edge features and the straight line features are simplified by using the Douglas–ρeucker algorithm; any straight line segment is selected and the length and direction of the straight line segment are calculated to obtain the lengths and directions of all straight line segments, all straight line segments in a preset direction interval [0, 180] are obtained, the preset direction interval is divided into a number of sub-direction intervals according to a preset step size, and the straight line eigenvalue ρ and the eigenvalue ρ in any sub-direction interval are calculated in turn. Where m is the number of straight lines in sub-direction interval j, n is the number of all sub-direction intervals, and qi is the length of any straight line segment; If there is any sub-direction interval whose straight line eigenvalue ρ is greater than a preset threshold, the sub-direction interval is marked as a target direction interval; the direction difference between the target direction intervals is calculated, and if there is a direction difference of 90 degrees, the edge area is marked as a background area; if there is no direction difference of 90 degrees between the target direction intervals, the edge area is marked as a transmission line area, a minimum circumscribed rectangle is circumscribed to the transmission line area, and the pixel rectangle is marked as the positioning frame of the transmission line.

3. The method for detecting surface defects of power equipment based on machine learning according to claim 2 is characterized in that: In S2, the infrared temperature of the transmission line is the highest temperature value in the transmission line area.

4. The method for detecting surface defects of power equipment based on machine learning according to claim 3 is characterized in that: The step S2 also includes screening the infrared image, and the specific screening process is as follows: Obtain the infrared temperature of all transmission lines in any infrared image, calculate the range C and temperature dispersion K corresponding to the infrared image, use the temperature dispersion as the horizontal coordinate and the range as the vertical coordinate, generate feature points corresponding to all infrared images, select any feature point as the center, set the clustering control radius R, obtain the feature point density ρ within the control radius R of any feature point, calculate the feature point density mean Avgρ, if there is any feature point with a feature point density ρ greater than Avgρ within the radius R, then use the feature point as the core point and generate a category cluster, generate several category clusters, obtain the number of feature points in any category cluster, select the category cluster with the largest number of feature points and mark it as the target category cluster, use the infrared image corresponding to the target category cluster as the target infrared image, obtain the infrared temperature of the transmission line and the positioning frame width in any target infrared image for function fitting; The specific calculation formula is: C = MAXT - MINT; Where n is the total number of transmission lines in the infrared image, MAXT is the maximum value of the transmission line temperature in the infrared image, and MINT is the minimum value of the transmission line temperature in the infrared image.

5. The method for detecting surface defects of power equipment based on machine learning according to claim 4 is characterized in that: The specific process of setting the cluster control radius R is: Taking any feature point as the center, calculate the Euclidean distance d between the feature point and any other feature point, sum each Euclidean distance to get u, and get the control radius R based on the value u. The calculation formula is as follows: Among them, u is the sum of the Euclidean distance data values ​​of all feature points, and d is the Euclidean distance between any feature points.

6. The method for detecting surface defects of power equipment based on machine learning according to claim 4, characterized in that: It also includes that if there is a non-core point in any category cluster and the density ρ within the control radius R is also greater than Avgρ, the category cluster generated by the non-core point is merged with the original category cluster to generate several category clusters.

7. The method for detecting surface defects of power equipment based on machine learning according to claim 1, characterized in that: In S4, the specific process of determining the target infrared image is as follows: S11, determine the standard positioning frame width D 标 , obtain the positioning frame width Da and infrared temperature Ta of any transmission line, and calculate D 标 The difference △H between the infrared temperature Ta of the transmission line and Da is substituted into the fitting function T(D) to calculate the fitting temperature T', and the corrected temperature Tb is obtained by summing the current infrared temperature Ta of the transmission line and the fitting temperature T'. S12, repeat the above step S11 to obtain the corrected temperature Tb of all transmission lines in any infrared image, and calculate the corrected temperature difference between any transmission line and all other transmission lines in turn. If any corrected temperature difference is greater than or equal to a preset threshold, the infrared image is calibrated as a target infrared image.

8. A detection system for surface defects of power equipment based on machine learning, characterized in that: include: The data acquisition module is used to set the target area, acquire the power transmission lines in the target area and mark them as the inspection route of the drone, and acquire the infrared monitoring video and optical monitoring video of the power transmission lines in real time through the drone; A data analysis module is used to perform frame processing on the infrared monitoring video according to a preset frame length to obtain an infrared image sequence y1, y2, ..., yN, select any infrared image and input it into a preset device recognition model to obtain a device image, generate a positioning frame of any transmission line from the device image, obtain the infrared temperature and positioning frame width of any transmission line, construct a point map with the positioning frame width as the horizontal coordinate and the infrared temperature as the vertical coordinate, fit the point map to obtain a fitting function T(D), and the fitting function T(D) is used to fit the relationship between the infrared temperature and the positioning frame width; The result generation module is used to correct the infrared temperature of any transmission line according to the fitting function T(D), determine the target infrared image according to the corrected infrared temperature, extract the optical image corresponding to the target infrared image and mark it as the target optical image, substitute the target optical image into the preset defect recognition model, and obtain the equipment defect recognition result of the transmission line.

Images