Image processing-based multi-type damage detection method and device for wind turbine blades

Abstract

The application discloses a wind turbine blade multi-type damage detection method and device based on image processing, comprising two processes of blade damage area identification and damage type judgment. The application firstly carries out image grayscale, filter enhancement, segmentation and morphological processing algorithms on the blade image collected by the intelligent inspection unmanned aerial vehicle platform, realizes the identification of the blade damage area; then obtains the geometric features and gray features and other information of the blade damage area based on the connected domain analysis principle; finally, designs a blade damage type identification classifier according to the extracted feature parameter information to accurately determine the blade damage types such as skin peeling, coating damage, sand eye, oil stain and crack. The application has the advantages of simple operation, low detection cost, high detection efficiency and the like, and makes up for the deficiencies of the traditional blade operation and maintenance detection method, such as long time consumption, large labor and material cost investment and the like.

Description

Technical Field

[0001] This invention belongs to the field of wind power generation technology, specifically relating to a method and device for detecting multiple types of damage to wind turbine blades based on image processing. Background Technology

[0002] As of the end of October 2022, my country's installed wind power capacity was approximately 350 million kilowatts, a year-on-year increase of 16.6%, maintaining a rapid growth trend. However, since my country began vigorously developing the wind power industry at the beginning of this century, the first batch of wind turbines has reached the end of their lifespan, and some have even exceeded their warranty period, leading to frequent failures. Blades account for a high proportion of the cost of wind turbines, and blade damage is one of the most frequent and costly types of component damage in wind turbine failures. This further highlights the importance and necessity of blade damage detection.

[0003] Currently, traditional methods for detecting damage to wind turbine blades mainly include high-magnification telescope inspection, blade maintenance platform inspection, and rope descent inspection (Spiderman). Although these methods can detect damage to blades to varying degrees, they require maintenance personnel to set up high-magnification telescopes, lifting platforms, and other equipment to conduct a comprehensive inspection of the blades. This results in various blind spots, high costs, low efficiency, high workload, long downtime, and personnel safety risks. Meanwhile, with the rapid development of computer vision technology, damage detection technology based on UAV inspection and image processing has become a research hotspot for wind turbine blade inspection. Its advantages are mainly reflected in: (1) lower inspection costs; (2) higher inspection efficiency; and (3) more systematic and comprehensive inspection content.

[0004] Therefore, in order to achieve automatic and accurate detection of various surface damages on wind turbine blades and reduce cost losses caused by blade damage, this invention proposes a method and device for detecting multiple types of damage to wind turbine blades based on image processing. Summary of the Invention

[0005] To address the shortcomings of traditional wind turbine blade damage detection methods, this invention provides a method and apparatus for detecting multiple types of damage to wind turbine blades based on image processing. The specific solution is as follows:

[0006] A method for detecting multiple types of damage to wind turbine blades based on image processing includes the following steps:

[0007] Step 1: Collect images of wind turbine blades according to the established inspection route and construct a wind turbine blade image set;

[0008] Step 2: Perform grayscale processing on the wind turbine blade image;

[0009] Step 3: Perform filtering and enhancement processing on the grayscale image;

[0010] Step 4: Perform segmentation processing on the filtered and enhanced leaf image to separate the foreground damaged area from the background area;

[0011] Step 5: Eliminate noisy pixels in the segmented image, fill in small holes, extract the complete leaf damage area, and complete the identification of the leaf damage area.

[0012] Step 6: Obtain the geometric features and grayscale feature parameters of the damaged area on the leaf;

[0013] Step 7: Based on the extracted geometric features and grayscale feature parameters, design a method for identifying the damage type of wind turbine blades, determine the type of blade damage, and complete the determination of the blade damage type.

[0014] To optimize the above technical solution, the specific measures also include:

[0015] In step one, the predetermined inspection route is as follows: the leaf to be inspected is locked in an inverted "Y" shape, and the image of a single leaf is acquired from the tip of the uppermost leaf to the root of the leaf. Then, the image acquisition of all leaves is completed in a counterclockwise order. During the acquisition process, the background area with light conditions below the threshold is illuminated.

[0016] In step two, the specific steps for grayscale processing of the wind turbine blade image are as follows: Using the average grayscale method, calculate the average pixel values ​​of the R, G, and B color channels of the color image, and set this average as the grayscale value of the grayscale image. The pixel calculation formula is:

[0017] Gray(i,j)=[R(i,j)+G(i,j)+B(i,j)] / 3 (1)

[0018] In the formula: R(i,j), G(i,j), B(i,j) represent the pixel values ​​of the three color channels R, G, and B, respectively; Gray(i,j) is the gray value of the resulting grayscale image; and i and j represent the row and column positions of the pixel in the image, respectively.

[0019] In step three, the specific method for filtering and enhancing the grayscale image employs median filtering and a contrast-limited adaptive histogram equalization enhancement algorithm. Median filtering replaces the pixel value at the filter center with the median value of all pixels within the filter range. The contrast-limited adaptive histogram equalization enhancement algorithm is based on the adaptive histogram equalization algorithm, pre-setting a threshold on the grayscale histogram, cutting off portions exceeding the threshold, and evenly distributing them across other grayscale levels. The expression for median filtering is:

[0020] g(x,y)=median{f(xi,yj),(i,j)∈G} (2)

[0021] In the formula: f(xi,yj) represents the pixel value after the filter template is shifted by i rows and j columns, g(x,y) represents the filtered pixel value, and G represents the two-dimensional filter template, which is usually a 3×3 or 5×5 region.

[0022] In step four, the filtered and enhanced leaf image is segmented using an image segmentation method. The image segmentation method combines the K-means algorithm and the Otsu algorithm to segment the enhanced image. The K-means clustering algorithm includes the following steps:

[0023] (1) Cluster the pixels in the image into K clusters and randomly generate K data centers;

[0024] (2) Calculate the distance from each pixel to each data center and assign it to the cluster corresponding to the data center with the shortest distance;

[0025] (3) Calculate the mean of each cluster and use the mean as the new data center;

[0026] (4) Repeat steps (2) and (3) until the termination condition of the iterative algorithm is met, and output the clustering results;

[0027] The clustering results are used as input to the Otsu thresholding algorithm to obtain the optimal threshold that maximizes the numerical variance between the segmented target region and the background region. The complete image segmentation map is then output according to this optimal threshold.

[0028] In step five, image morphology processing methods are used to eliminate noisy pixels and fill small holes in the segmented image. Specifically, the image morphology processing method involves: first performing a dilation operation on the segmented image using a closing operation, then performing an erosion operation to fill the small holes inside the segmented image, connecting adjacent regions, and then setting a minimum connected region area threshold. If the area is less than this threshold, it is identified as a noisy pixel and eliminated, thereby extracting the complete leaf damage area and completing the identification of the leaf damage area. The expression for the closing operation is:

[0029]

[0030] In the formula: A is the target image to be processed; B is the structuring element used for the operation; · is the closing operator; Θ is the symbol for the erosion operation; The symbol for the expansion operation.

[0031] In step six, the geometric features and grayscale feature parameters of the leaf damage area are obtained based on the image connected component analysis. The specific method is as follows: pixels with the same pixel value and adjacent positions in the image are grouped into several regions. All pixels in the same region are marked with the same value or symbol, and the pixels in different regions are marked with different values ​​or symbols. The geometric features and grayscale features of the leaf damage area are obtained based on the regions. The geometric features of the leaf damage area are: area, perimeter, length, width, area of ​​the minimum bounding rectangle, ratio of length to minor diameter, rectangle similarity, shape factor, elongation, and degree of damage. The grayscale features are the mean and variance of the grayscale of the leaf damage area.

[0032] In step seven, a method for identifying the damage type of wind turbine blades is designed based on the extracted geometric features and grayscale feature parameters. The specific method for determining the blade damage type is as follows: Let the area of ​​the target image be represented by the symbol S, and the size be M*N; the image processing result image is a black background with a white foreground, and black and white pixels are represented by 0 and 1 respectively. G(x,y) is the grayscale value at the (x,y) coordinates. The minimum bounding rectangle is the smallest rectangle that encloses the damage area, and its four vertices are a, b, c, and d respectively. The extracted geometric features and grayscale feature parameters of the blade damage area are as follows:

[0033] 1) The area A1 is calculated using the following formula:

[0034]

[0035] 2) Perimeter C, the formula for calculation is:

[0036]

[0037] ΔC is the length of the line connecting any two adjacent pixels on the boundary of the blade damage area;

[0038] 3) Length L, the calculation formula is:

[0039] L = max{L ab ,L bc} (6)

[0040] L ab and L bc These represent the lengths between points a and b, and the lengths between points b and c, respectively.

[0041] 4) Width W, calculated using the following formula:

[0042] W = min{L ab ,L bc} (7)

[0043] 5) The area A2 of the minimum circumscribed rectangle is calculated using the following formula:

[0044] A2=L×W (8)

[0045] 6) The ratio of the major to minor diameter, R, is calculated using the following formula:

[0046]

[0047] 7) Rectangle similarity D, calculated using the following formula:

[0048]

[0049] 8) Shape factor F, calculated using the following formula:

[0050]

[0051] 9) The elongation length T is calculated using the following formula:

[0052]

[0053] 10) Damage degree ε, calculated using the following formula:

[0054]

[0055] 11) The grayscale mean GM is calculated using the following formula:

[0056]

[0057] Gray i (x,y) represents the gray value within the damaged area;

[0058] 12) Gray-level variance (GV) is calculated using the following formula:

[0059]

[0060] After extracting the geometric and grayscale feature parameters of the damaged leaf area, the first step is to determine whether the shape factor F is less than the shape factor threshold F. lim If the area is less than the threshold value, it is determined to be trachoma damage; otherwise, it is determined again whether the area of ​​the minimum bounding rectangle A2 is greater than the threshold value A. 2lim If the ratio is greater than 1, it is determined to be an epidermal detachment injury; otherwise, it is determined again whether the ratio of the major and minor axes R is greater than the ratio of the major and minor axes R. lim If the value is greater than GM, it is determined to be crack damage; otherwise, the mean gray value GM and the variance gray value are checked again to see if they meet the following conditions: GM lim1 <GM<GM lim2 GV lim1 <GV<GV lim2 Among them, GM lim1 and GM lim2 These represent the minimum and maximum thresholds for the mean grayscale value, respectively.lim1 and GV lim2 These represent the minimum and maximum thresholds of the grayscale variance, respectively. If the above conditions are met, the damage is determined to be caused by oil stains; otherwise, it is determined to be caused by coating damage.

[0061] The image processing-based wind turbine blade multi-type damage detection device includes an intelligent inspection drone platform, an image grayscale processing module, a filtering and enhancement processing module, a segmentation processing module, an image morphology processing module, an image connected component analysis module, and a wind turbine blade damage type identification and classification module. These modules are sequentially connected by signals.

[0062] The intelligent inspection drone platform is used to collect images of wind turbine blades according to a predetermined inspection route and build a wind turbine blade image set.

[0063] The image grayscale processing module is used to perform grayscale processing on the images of wind turbine blades;

[0064] The filtering and enhancement processing module is used to perform filtering and enhancement processing on grayscale images;

[0065] The segmentation processing module is used to segment the filtered and enhanced leaf image to separate the foreground damaged area from the background area.

[0066] The image morphology processing module is used to eliminate noisy pixels in the segmented image, fill small holes, extract complete leaf damage areas, and complete the identification of leaf damage areas.

[0067] The image connected component analysis module is used to obtain the geometric features and grayscale feature parameters of the damaged area of ​​the leaf;

[0068] The wind turbine blade damage type identification and classification module is used to determine the blade damage type based on the extracted geometric features and grayscale feature parameters.

[0069] The intelligent inspection drone platform includes an inspection drone, a high-definition camera, an image acquisition card, and a stabilization gimbal. The stabilization gimbal is mounted on the inspection drone, the high-definition camera is rotatably mounted on the stabilization gimbal, and the image acquisition card is mounted on the high-definition camera.

[0070] The inspection drone is also equipped with a lighting module, which is used to illuminate background areas where the light conditions are below a threshold.

[0071] The lighting module includes interconnected light sources and a lighting controller, which controls the light emitted by the light source.

[0072] The beneficial effects of this invention are as follows:

[0073] (1) This invention uses an intelligent inspection drone platform to perform wind turbine blade image acquisition according to a predetermined inspection route to construct a wind turbine blade image set; it uses image grayscale conversion to process the wind turbine blade images to reduce the amount of data and improve the algorithm's computational efficiency; it uses image filtering and image enhancement methods to filter and enhance the grayscale images to reduce image noise and improve the recognizability of damaged areas, enhancing the detailed features in the blade images; it uses image segmentation methods to segment the filtered and enhanced blade images to achieve the segmentation of the foreground damaged area and the background area; it uses image morphology processing methods to eliminate noise pixels in the segmented images and fill small holes to extract complete blade damaged areas, thus completing the identification of blade damaged areas; it obtains the geometric features and grayscale features of the blade damaged areas based on the principle of image connected component analysis; and it designs a wind turbine blade damage type identification classifier based on the extracted feature parameters to accurately determine blade damage types such as skin peeling, coating damage, sand holes, oil stains, and cracks, thus completing the blade damage type judgment. This invention is driven by the need for intelligent diagnosis of the health status of wind turbine blades in my country's wind power new energy field. It studies and solves the problem of detecting multiple types of surface damage on wind turbine blades under the new situation of digital empowerment. It not only meets the needs of the wind power industry for cost reduction and efficiency improvement, but also facilitates the intelligent operation and maintenance of wind farms. It has a good application prospect in the operation and maintenance of wind turbine blades.

[0074] (2) Compared with traditional blade maintenance and inspection methods: This invention no longer relies on large equipment such as lifting platforms and excessive labor costs; This invention is a non-contact damage detection method, which cleverly utilizes the high degree of automation and simple operation of intelligent inspection drone platforms to complete image acquisition. The designed method can efficiently and quickly identify the damaged area of ​​the blade and determine the damage type; It has the advantages of less downtime for wind turbine units, higher degree of automation, and higher detection efficiency, and has strong practicality and broad application prospects.

[0075] (3) Compared with similar image processing-based blade damage detection methods: This invention combines K-means and Otsu algorithm to improve image segmentation accuracy. The segmentation effect image has fewer noise and internal voids, and the blade damage contour is smoother. The improved closing operation method in this invention realizes the effective filling of small voids in the segmentation image and the accurate elimination of noise pixels, and completely extracts the blade damage area, laying the foundation for the accurate acquisition of wind turbine blade damage feature information. This invention is not only for single damage type detection, but also designs a wind turbine blade damage type identification classifier based on the extracted geometric features such as area, perimeter, length, width, area of ​​minimum bounding rectangle, ratio of length to minor diameter, rectangle similarity, shape factor, elongation and damage degree of the blade damage area, as well as gray features such as gray mean and gray variance of the blade damage area. It realizes the judgment of blade damage such as skin peeling, coating damage, sand hole, oil stains and cracks, and provides timely and accurate reference for wind turbine blade fault detection and maintenance. Attached Figure Description

[0076] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0077] Figure 1 This is a flowchart of the image processing-based method for detecting multiple types of damage to wind turbine blades according to the present invention.

[0078] Figure 2 This is a schematic diagram of the detection system in operation according to an embodiment of the present invention.

[0079] Figure 3 These are images of leaf samples collected by the intelligent inspection drone platform in this embodiment of the invention.

[0080] Figure 4 This is a grayscale effect diagram obtained by the average grayscale conversion method in an embodiment of the present invention.

[0081] Figure 5 This is a diagram showing the filtering effect obtained by the median filtering method in an embodiment of the present invention.

[0082] Figure 6 This is an image showing the enhancement effect obtained by the CLAHE image enhancement method according to an embodiment of the present invention.

[0083] Figure 7 This is a segmentation result obtained by combining K-means and Otsu's algorithm in an embodiment of the present invention.

[0084] Figure 8 This is a morphological processing effect diagram obtained by the improved closing operation method in an embodiment of the present invention.

[0085] Figure 9 This is a flowchart illustrating the workflow of a wind turbine blade damage type identification and classifier according to an embodiment of the present invention. Detailed Implementation

[0086] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0087] The flowchart of a multi-type damage detection method for wind turbine blades based on image processing according to the present invention is as follows: Figure 1 As shown, the overall damage detection method includes the following steps:

[0088] Step 1: The inspection drone, equipped with a high-definition camera, image acquisition card, and stabilization gimbal, performs blade image acquisition according to a predetermined inspection route. For background areas with poor lighting conditions, an illumination module is used to supplement the light. Then, the blade images are transmitted via a data transmission module for use as input for subsequent image processing and analysis. A schematic diagram of the inspection system is shown below. Figure 2 As shown, the collected leaf sample images are as follows: Figure 3 As shown.

[0089] Step 2: Use the average grayscale method to perform image grayscale processing to reduce the amount of data and improve the system's computational efficiency. The grayscale effect is shown in the image below. Figure 4 As shown, the average grayscale pixel value is calculated according to the following formula:

[0090] Gray(i,j)=[R(i,j)+G(i,j)+B(i,j)] / 3 (1)

[0091] In the formula: R(i,j), G(i,j), B(i,j) represent the pixel values ​​of the three color channels R, G, and B, respectively, and Gray(i,j) is the gray value of the grayscale image obtained after conversion.

[0092] Step 3: Apply median filtering to the image to remove noise caused by factors such as weather, uneven lighting, and the intelligent inspection drone platform itself. The filtering effect is shown in the image below. Figure 5 As shown, median filtering is calculated according to the following formula:

[0093] g(x,y)=median{f(xi,yj),(i,j)∈G} (2)

[0094] In the formula: f(x,y) and g(x,y) represent the pixel values ​​before and after filtering, and G represents the two-dimensional filtering template, which is usually a 3×3 or 5×5 region.

[0095] Step 4: Apply the contrast-limited adaptive histogram equalization algorithm (CLAHE algorithm) to improve the recognizability of the damaged area and enhance the detailed features in the leaf image; the resulting enhanced image is shown below. Figure 6 As shown.

[0096] Step 5: Perform initial image segmentation using the K-means algorithm, followed by a second segmentation using the Otsu's algorithm to separate the damaged area from the background area. The K-means clustering algorithm mainly includes the following four steps: 1) Cluster the pixels in the image into K clusters and randomly generate K data centers; 2) Calculate the distance from each pixel to each data center and assign it to the cluster corresponding to the data center with the shortest distance; 3) Calculate the mean of each cluster and use the mean as the new data center; 4) Repeat steps 2 and 3 until the termination condition of the iterative algorithm is met (reaching the maximum number of iterations or the change in cluster centers meets the allowable error), and output the clustering result. The resulting segmentation effect is shown in the image below. Figure 7 As shown.

[0097] Step Six: Initially extract the damaged area of ​​the leaf using a closing operation. Then, set a minimum connected component area threshold and determine if the connected component area in the closed operation processing image is less than this threshold. If it is less than the threshold, it is identified as a noisy pixel and eliminated to extract a more complete leaf damage area, thus achieving leaf damage area identification. The resulting morphological processing image is shown below. Figure 8 As shown, the closing operation is calculated according to the following formula:

[0098]

[0099] In the formula: A is the target image to be processed; B is the structuring element used for the operation; · is the closing operator; Θ is the symbol for the erosion operation; The symbol for the expansion operation.

[0100] Step 7: Extract geometric features and grayscale features of the damaged area based on the principle of connected component analysis. The feature information is further explained and described in more detail below: Let the area of ​​the target image be represented by the symbol S, and the size be M*N; the image processing result image has a black background and a white foreground, with black and white pixels represented by 0 and 1 respectively, and G(x,y) being the grayscale value at the (x,y) coordinates. The minimum bounding rectangle is the smallest rectangle that encloses the damaged area, and its four vertices are a, b, c, and d. The feature information of the wind turbine blade damage area to be extracted is shown in Table 1.

[0101] Table 1. Characteristics of Damage Areas in Wind Turbine Blades

[0102]

[0103] Step 8: Input the extracted blade damage feature parameters into the blade damage type identification classifier to accurately determine the blade damage type, such as skin peeling, coating damage, pinholes, oil stains, and cracks. The workflow diagram of the wind turbine blade damage type identification classifier is shown below. Figure 9 As shown in the figure: F and F lim Representing the shape factor and its threshold respectively; A2 and A 2lim Represent the area of ​​the minimum bounding rectangle and its threshold, respectively; R and R lim These represent the ratio of the major and minor axes and their threshold values, respectively; GM, GM lim1 and GM lim2 These represent the mean grayscale value and its minimum and maximum threshold values, respectively; GV, GV lim1 and GV lim2 These represent the grayscale variance and its minimum and maximum thresholds, respectively.

[0104] The wind turbine blade multi-type damage detection device based on image processing of the present invention includes an intelligent inspection drone platform, an image grayscale processing module, a filtering and enhancement processing module, a segmentation processing module, an image morphology processing module, an image connected component analysis module, and a wind turbine blade damage type identification and classification module. The intelligent inspection drone platform, image grayscale processing module, filtering and enhancement processing module, segmentation processing module, image morphology processing module, image connected component analysis module, and wind turbine blade damage type identification and classification module are sequentially signal-connected.

[0105] The intelligent inspection drone platform is used to collect images of wind turbine blades according to a predetermined inspection route and build a wind turbine blade image set.

[0106] The image grayscale processing module is used to perform grayscale processing on the images of wind turbine blades;

[0107] The filtering and enhancement processing module is used to perform filtering and enhancement processing on grayscale images;

[0108] The segmentation processing module is used to segment the filtered and enhanced leaf image to separate the foreground damaged area from the background area.

[0109] The image morphology processing module is used to eliminate noisy pixels in the segmented image, fill small holes, extract complete leaf damage areas, and complete the identification of leaf damage areas.

[0110] The image connected component analysis module is used to obtain the geometric features and grayscale feature parameters of the damaged area of ​​the leaf;

[0111] The wind turbine blade damage type identification and classification module is used to determine the blade damage type based on the extracted geometric features and grayscale feature parameters.

[0112] The intelligent inspection drone platform includes an inspection drone, a high-definition camera, an image acquisition card, and a stabilization gimbal. The stabilization gimbal is mounted on the inspection drone, the high-definition camera is rotatably mounted on the stabilization gimbal, and the image acquisition card is mounted on the high-definition camera.

[0113] The inspection drone is also equipped with a lighting module, which is used to illuminate background areas where the light conditions are below a threshold.

[0114] The lighting module includes interconnected light sources and a lighting controller, which controls the light emitted by the light source.

[0115] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for detecting multiple types of damage to wind turbine blades based on image processing, characterized in that, Includes the following steps: Step 1: Collect images of wind turbine blades according to the established inspection route and construct a wind turbine blade image set; Step 2: Perform grayscale processing on the wind turbine blade image; Step 3: Perform filtering and enhancement processing on the grayscale image; Step 4: Perform segmentation processing on the filtered and enhanced leaf image to separate the foreground damaged area from the background area; Step 5: Eliminate noisy pixels in the segmented image, fill in small holes, extract the complete leaf damage area, and complete the identification of the leaf damage area. Step 6: Obtain the geometric features and grayscale feature parameters of the damaged area on the leaf; Step 7: Design a method for identifying the damage type of wind turbine blades based on the extracted geometric features and grayscale feature parameters, determine the type of blade damage, and complete the determination of the blade damage type. In step seven, a method for identifying wind turbine blade damage types is designed based on the extracted geometric features and grayscale feature parameters. The specific method for determining the blade damage type is as follows: Let the area of ​​the target image be represented by the symbol S, and the size be M*N; the image processing result image is a black background with a white foreground, and black and white pixels are represented by 0 and 1 respectively. I(x,y) is the grayscale value at the (x,y) coordinate. The minimum bounding rectangle is the smallest rectangle that encloses the damage area, and its four vertices are a, b, c, and d respectively. The extracted geometric features and grayscale feature parameters of the blade damage area are as follows: 1) Area The calculation formula is: （4）； 2) Perimeter C, the formula for calculation is: （5）； The length of the line connecting any two adjacent pixels on the boundary of the blade damage area; 3) Length L, the formula for calculation is: （6）； These represent the lengths between points a and b, and the lengths between points b and c, respectively. 4) Width W, calculated using the following formula: （7）； 5) Area of ​​the minimum circumscribed rectangle The calculation formula is: （8）； 6) The ratio of the major to minor diameter, R, is calculated using the following formula: （9）； 7) Rectangle Similarity The calculation formula is: （10）； 8) Shape factor The calculation formula is: （11）； 9) Elongation The calculation formula is: （12）； 10) Degree of damage The calculation formula is: （13）； 11) Gray-scale mean The calculation formula is: （14）； Represents the grayscale value within the damaged area; 12) Gray-scale variance The calculation formula is: （15）； After extracting the geometric and grayscale feature parameters of the damaged leaf area, the first step is to determine whether the shape factor F is less than the shape factor threshold F. lim If it is less than , it is determined to be trachoma damage; Conversely, if the minimum bounding rectangle area A2 is greater than the minimum bounding rectangle area threshold A, then check again. 2lim If the ratio is greater than 1, it is determined to be an epidermal detachment injury; otherwise, it is determined again whether the ratio of the major and minor axes R is greater than the threshold R. lim If the value is greater than GM, it is determined to be crack damage; otherwise, the mean gray value GM and the variance gray value are checked again to see if they meet the following conditions: GM lim1 <GM<GM lim2 GV lim1 <GV<GV lim2 Among them, GM lim1 and GM lim2 These represent the minimum and maximum thresholds for the mean grayscale value, respectively. lim1 and GV lim2 These represent the minimum and maximum thresholds of grayscale variance, respectively. If the above conditions are met, the damage is determined to be caused by oil stains; otherwise, it is determined to be caused by coating damage.

2. The method for detecting multiple types of damage to wind turbine blades based on image processing according to claim 1, characterized in that, In step one, the predetermined inspection route is as follows: the leaf to be inspected is locked in an inverted "Y" shape, and the image of a single leaf is acquired from the tip of the uppermost leaf to the root of the leaf. Then, the image acquisition of all leaves is completed in a counterclockwise order. During the acquisition process, the background area with light conditions below the threshold is illuminated.

3. The method for detecting multiple types of damage to wind turbine blades based on image processing according to claim 1, characterized in that, In step two, the specific steps for grayscale processing of the wind turbine blade image are as follows: Using the average grayscale method, calculate the average pixel values ​​of the R, G, and B color channels of the color image, and set this average as the grayscale value of the grayscale image. The pixel calculation formula is: （1）； In the formula: R(i,j), G(i,j), B(i,j) represent the pixel values ​​of the three color channels R, G, and B, respectively; Gray(i,j) is the gray value of the resulting grayscale image; and i and j represent the row and column positions of the pixel in the image, respectively.

4. The method for detecting multiple types of damage to wind turbine blades based on image processing according to claim 1, characterized in that, In step three, the specific method for filtering and enhancing the grayscale image employs median filtering and a contrast-limited adaptive histogram equalization enhancement algorithm. Median filtering replaces the pixel value at the filter center with the median value of all pixel values ​​within the filter range. The contrast-limited adaptive histogram equalization enhancement algorithm is based on the adaptive histogram equalization algorithm, pre-setting a threshold on the grayscale histogram, cutting off portions exceeding the threshold, and evenly distributing them across other grayscale levels. The expression for median filtering is: （2）； In the formula: f(xi,yj) represents the pixel value after shifting the filter template by i rows and j columns, g(x,y) represents the filtered pixel value, and J represents the two-dimensional filter template, which is usually... area.

5. The method for detecting multiple types of damage to wind turbine blades based on image processing according to claim 1, characterized in that, In step four, an image segmentation method is used to segment the filtered and enhanced leaf image. This image segmentation method combines the K-means algorithm and the Otsu's algorithm to segment the enhanced image. The K-means algorithm includes the following steps: (1) Cluster the pixels in the image into K clusters and randomly generate K data centers; (2) Calculate the distance from each pixel to each data center and assign it to the cluster corresponding to the data center with the shortest distance; (3) Calculate the mean of each cluster and use the mean as the new data center; (4) Repeat steps (2) and (3) until the termination condition of the iterative algorithm is met, and output the clustering results; The clustering results are used as input to the Otsu thresholding algorithm to obtain the optimal threshold that maximizes the numerical variance between the segmented target region and the background region. The complete image segmentation map is then output according to this optimal threshold.

6. The method for detecting multiple types of damage to wind turbine blades based on image processing according to claim 1, characterized in that, In step five, image morphological processing methods are used to eliminate noisy pixels and fill small holes in the segmented image. Specifically, the image morphological processing method involves: performing a dilation operation on the segmented image using a closing operation, followed by an erosion operation to fill the small holes inside the segmented image, connecting adjacent regions, and then setting a minimum connected region area threshold. If the area is less than this threshold, it is identified as a noisy pixel and eliminated, thereby extracting the complete leaf damage area and completing the identification of the leaf damage area. The expression for the closing operation is: （3）； In the formula: A is the target image to be processed; B is the structuring element used for the calculation; It is the closing operator; The symbol for the erosion operation; The symbol for the expansion operation.

7. The method for detecting multiple types of damage to wind turbine blades based on image processing according to claim 1, characterized in that, In step six, the geometric features and grayscale feature parameters of the leaf damage area are obtained based on image connected component analysis. Specifically, the method is as follows: pixels with the same pixel value and adjacent positions in the image are grouped into several regions. All pixels in the same region are marked with the same value or symbol, and the pixel values ​​or symbols used in different regions are different. The geometric features and grayscale features of the leaf damage area are obtained based on the regions. The geometric features of the leaf damage area are: area, perimeter, length, width, area of ​​the minimum bounding rectangle, ratio of major to minor diameter, rectangle similarity, shape factor, elongation, and degree of damage. The grayscale features are the grayscale mean and grayscale variance of the leaf damage area.

Images