Offshore wind turbine blade defect detection method based on image enhancement and Hough transformation

Abstract

The invention discloses an offshore wind turbine blade defect detection method based on image enhancement and Hough transformation, and the method comprises the following steps: carrying out the enhancement of an offshore wind turbine blade image shot by an unmanned plane, and reducing the overexposure phenomenon caused by light; the fan blade is positioned based on Hough transformation; and identifying the surface defects of the fan blade based on the OTSU. The improved Hough transformation and pixel tracking edge detection method is adopted to recognize the fan blade, the pixel point field is divided into eight adjacent areas, pixel filling operation is carried out, interference of objects with linear features such as cloud layers can be reduced, and the detection precision of the fan blade is improved. According to the method, the image is divided into a foreground part and a background part through OTSU binarization segmentation, the corrosion condition of the surface of the fan blade can be obtained in a non-contact mode, and extra damage to a measured object is avoided.

Description

Technical Field

[0001] The present invention relates to the field of defect detection of offshore wind turbine blades, and specifically relates to a method for defect detection of offshore wind turbine blades based on image enhancement and Hough transform. Background Art

[0002] After a wind turbine blade has been operating for some time, it will be affected by the natural environment and show varying degrees of surface corrosion, which will in turn affect the power generation efficiency. In severe cases, it may even cause the blade to break. Especially for offshore wind farms, which are under complex marine climate conditions and are often affected by salt spray, lightning strikes, and rain erosion, corrosion phenomena frequently occur. Therefore, it is urgent to focus on the corrosion defects of wind turbine blades in order to timely repair the blades. However, when a drone collects images of wind turbine blades with white characteristics and certain reflectivity in a marine environment where the colors are mainly blue and white and the lighting conditions change drastically, it is often easy to cause local "bright spots" or "overexposure" phenomena, which will affect the extraction effect of the blade contour and defect features by traditional image processing techniques, and even result in missed detections. How to enhance the image and accurately identify the wind turbine blade is a current technical problem.

[0003] Especially for offshore wind farms that are under complex marine climates for a long time and are often affected by salt spray, lightning strikes, and rain erosion, resulting in the corrosion of wind turbine blades, it is necessary to identify the corrosion. When a drone takes pictures of offshore wind turbine blades, it is easily affected by the surrounding environment, resulting in local bright spots or overexposure in the captured pictures, which will interfere with the positioning and corrosion detection of wind turbine blades and even result in missed detections. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a method for defect detection of offshore wind turbine blades based on image enhancement and Hough transform. First, the overexposure problem of the captured pictures is solved through image enhancement, and then the wind turbine blades in the image are located based on the improved Hough transform. Finally, the corrosion defects are accurately identified by using OTSU threshold segmentation and combining with the surface roughness calculation method.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A method for defect detection of offshore wind turbine blades based on image enhancement and Hough transform, comprising the following steps:

[0007] Step 1: Enhance the image of the offshore wind turbine blade captured by the drone to reduce the overexposure phenomenon caused by light.

[0008] Step 2: Locate the wind turbine blade based on the Hough transform;

[0009] Step 3: Identify the surface defects of the fan blade based on OTSU.

[0010] The said step 1 includes the following steps:

[0011] S1.1: Convert the fan blade image from the RGB space to the HSV (Hue, Saturation, Value) space, and extract the V channel;

[0012] S1.2: Adjust the contrast and exposure of the image using the CLAHE algorithm within the V channel;

[0013] S1.3: Modify the probability density function (pdf) and cumulative distribution function (cdf) of the image, and adjust the local mean coefficient K to improve the detail correction effect of the local scene.

[0014] In S1.1, convert the fan blade image from the RGB space to the HSV (Hue, Saturation, Value) space, and extract the V channel; specifically as follows:

[0015] 1): Map the R, G, B components of each pixel of the image to the interval (0, 1), and the formula is as follows:

[0016]

[0017] Among them, R, G, B respectively represent the values of the original pixel channels of the image, 255 is the value range of each channel pixel value, and R’, G’, B’ respectively represent the pixel values of each channel after normalization, and their range interval is (0, 1).

[0018] 2): Directly extract the V channel. The V channel represents the brightness information of the image, and its value is the maximum value among R’, G’, B’. The formula is as follows:

[0019] V = max(R', G', B') (2);

[0020] In S1.2, adjust the contrast and exposure of the image using the CLAHE algorithm within the V channel; specifically as follows:

[0021] 1) Divide the V channel image into multiple 8×8 sub-blocks, and perform mirror padding on the image edges. The formula is as follows:

[0022]

[0023] Among them, V(x, y) represents the V-channel value of the image at the pixel point (x, y), V’(x, y) represents the V-channel value of the filled image, and Width and Height represent the width and height of the original image respectively.

[0024] 2) Calculate the local histogram and limit the contrast:

[0025]

[0026] Among them, k is the gray level, L is the total number of gray levels, α is the contrast limit factor, h(k) represents the original frequency of gray level k in the sub-block, h clip (k) is the frequency of gray level k after clipping, and h final is the final frequency after compensation.

[0027] 3) Histogram equalization:

[0028]

[0029] T(k) is the gray mapping function, L is the total number of gray levels, N is the total number of pixels in the sub-block, and h final is the final frequency after compensation.

[0030] 4) Bilinear interpolation to merge sub-blocks:

[0031]

[0032] Among them, 4 is the number of adjacent sub-blocks of the current pixel, including the upper left, upper right, lower left, and lower right; γ i is the weight, which is determined by the distance from the pixel to the sub-block; T i is the gray mapping function of the i-th sub-block; V’(x, y) represents the V-channel value of the filled image, and V out (x, y) is the final output V-channel value.

[0033] S1.3 includes the following steps:

[0034] 1): Based on the statistical distribution of the image gray level hierarchy, adopt the average strategy to achieve the purpose of histogram equalization, so as to correct the probability density function of the image. The formula is as follows:

[0035]

[0036] Among them, pdf n (l) represents the corrected probability density function; pdf(l) represents the current image gray level distribution; pdf max and pdf min represent the set upper and lower limits of the probability distribution; pdf cow represents the compensation component; ε represents the proportional reduction factor, 0 < ε < 1.

[0037] 2): Correct the cumulative distribution function of the image, and the formula is as follows:

[0038]

[0039] where cdf Ω (l) represents the value of the cumulative distribution function of the gray level l in the corrected window area.

[0040] 3): Set a local mean coefficient K to adjust the gamma coefficient. This value determines whether the gamma value is greater than 1. Finally, obtain the output O(l) after local adaptive gamma correction, and the formula is as follows:

[0041]

[0042] where c is a constant, 0 < c < 1; represents the average brightness in the current local window area; Ω represents the local window; l represents the brightness value of the central pixel of the local window after histogram equalization processing; l max represents the maximum value of the image gray level; O(l) represents the output brightness value after local adaptive gamma correction;

[0043] The step 2 includes the following steps:

[0044] S2.1. Object detection based on the pixel tracking edge extraction method:

[0045] First, select local anchor points as the starting points for detection. Qualified local anchor points need to meet the following two conditions simultaneously:

[0046] 1) Have the same gradient direction as the adjacent pixels in the edge direction;

[0047] 2) Have the maximum inner product energy value among the adjacent pixels in the pixel gradient direction.

[0048] The selection process is as Figure 3 shown. The numbers marked in parentheses indicate the inner product energy values of the pixel points. The solid arrows are responsible for indicating the gradient directions of the pixels, and the dashed arrows are responsible for indicating the edge directions of the pixels. In Figure 3 subfigure a, the gradient directions of the pixels in the neighborhood of point X in the edge direction are inconsistent, so it is determined as an unqualified anchor point. In Figure 3 subfigure b, the inner product energy value of point X in the gradient direction is not the highest, and it is also recognized as an unqualified anchor point. Only in Figure 3 subfigure c, point X meets the above two conditions simultaneously and is confirmed as a qualified anchor point.

[0049] Subsequently, connect each edge pixel one by one along the image edge direction;

[0050] The initial anchor point connects the discrete edge points from the left and right directions using the edge expansion direction. The tracking direction of subsequent edge pixel points is determined by the edge direction of the edge pixel points. The tracking direction of subsequent edge pixel points is determined by the edge direction of the edge pixel points. Edge pixel detection needs to meet the following conditions:

[0051] 1) The inner product energy amplitude of the edge point is the largest;

[0052] 2) The edge point is a maximum point in its neighborhood;

[0053] 3) The edge direction of the edge point is the same as that of the previous edge point.

[0054] The detection process is as Figure 4 shown. In sub - figure a of Figure 4 , first, the edge tracking direction is divided into 8 parts, and the specific path of edge tracking in each part is determined by the direction - encoding value c. Figure 4 Sub - figure b of

[0055] shows the search area of the next edge pixel point. A0 represents the previous edge pixel point, and A1 represents the current edge pixel point. If the given direction - encoding value of the current edge pixel point is c, then the search area of the subsequent edge pixel point can be accurately defined according to the rule {c, (c + 1)&0x7, (c - 1)&0x7}.

[0056] Finally, a series of smooth curves are constructed;

[0056] Candidate pixel points that meet specific three criteria will be evaluated for confidence according to a preset weight. The criteria for confidence evaluation are as follows:

[0057] 1) Calculate the gradient magnitude G(x, y):

[0058]

[0059] where G x represents the gradient in the horizontal direction, and G y represents the gradient in the vertical direction.

[0060] 2) Direction consistency D(c, c’):

[0061] D(c, c') = cos(θ(c)-θ(c')) (12);

[0062] where c is the current direction encoding, c’ is the candidate direction encoding, and θ(c) and θ(c’) are the angles corresponding to the direction encodings respectively.

[0063] 3) Neighborhood continuity N(x, y):

[0064]

[0065] Among them, the neighborhood window size is 3×3.

[0066] 4) Calculate the confidence Confidence(x, y) by synthesizing the above three criteria:

[0067]

[0068] Among them, ω1, ω2, and ω3 are weights, satisfying ω1 + ω2 + ω3 = 1, and G max is the maximum gradient amplitude of the image.

[0069] Determine the pixel point with the highest confidence as the final edge pixel point, and make corresponding adjustments to the direction of edge tracking. If no continuous edge pixel points meeting the conditions can be detected in a specific direction, it can be inferred that the edge tracking process in this direction has reached the termination condition; at this time, the tracking process will turn to another direction to continue performing edge detection. By fusing the sets of edge pixel points obtained in two independent directions, the complete edge contour of the image is constructed.

[0070] S2.2. Target localization based on improved Hough line detection:

[0071] According to the edge point position information, divide the pixel point neighborhood into eight adjacent regions and perform pixel filling operations. After obtaining the line pixel points, fit all the pixel point information belonging to the same line by the least squares method.

[0072] S2.2 includes the following steps:

[0073] S2.2.1: According to the edge point position information, divide the pixel point neighborhood into eight adjacent regions and perform pixel filling operations;

[0074] S2.2.2: Calculate the distance from the pixel points belonging to the line to the line and set a threshold. The specific steps are as follows:

[0075] 1) The straight line equation detected by the Hough transform is:

[0076] y = ax + b (15);

[0077] Among them, a and b represent the slope and intercept of the straight line respectively.

[0078] 2) Calculate the distance d from the pixel point to the straight line i :

[0079]

[0080] Among them, x i and y i represent the horizontal and vertical coordinates of the pixel point respectively, and a and b represent the slope and intercept of the straight line respectively as above.

[0081] 3) Set the distance threshold T:

[0082] T = k·σ (17);

[0083] Where k is the adjustment coefficient and σ is the standard deviation of the noise. The calculation formula is as follows:

[0084]

[0085] Where the pixel values extracted from the pure color area are recorded as the set {v1, v2,..., v n}, and μ is the mean value of the pixels in this area.

[0086] S2.2.3: Judge the relationship between the distance value d i and the threshold T. When the distance value is less than the threshold, it is determined as the pixel of the straight line point and included in the least squares fitting to improve the fitting accuracy. The fitting process is as follows:

[0087]

[0088] Where (x i , y i ) represents the position of the straight line pixel point, x i , y i respectively represent the horizontal and vertical coordinates of the pixel point, and a and b represent the straight line parameters to be fitted; and represent the best parameter estimation values; W represents the error function of the straight line fitting, and N is the number of pixel points.

[0089] Step 3 includes the following steps:

[0090] S3.1. Segment the image into foreground and background parts based on OTSU-based binary segmentation; specifically as follows:

[0091] 1) Construct a two-dimensional histogram, and the formula is as follows:

[0092]

[0093] Where i and j respectively represent the gray level of the pixel point (x, y) and the average gray value within the 3×3 neighborhood centered on the pixel point (x, y), and P ij is the probability that the number of pixels with gray level i and neighborhood mean value j appears.

[0094] 2) Define the weight and the mean value:

[0095] Let the probabilities of the background and foreground appearance be λ1 and λ2 respectively, the corresponding mean vectors be μ1 and μ2 respectively, the mean vector corresponding to the entire image be μ, s and t be the thresholds within the ranges of i and j respectively, and the relevant calculation formulas are as follows:

[0096]

[0097] μ = (∑iP ij , ∑jP ij ) (27);

[0098] 3) Calculate the trace of the between-class variance and select the optimal threshold. The formula is as follows:

[0099] tr(S b ) = λ1[(μ 1i - μ i ) 2 + (μ 1j - μ j ) 2 + λ2[(μ 2i - μ i ) 2 + (μ 2j - μ j ) 2 (28);

[0100]

[0101] Among them, S b is the discrete measure matrix, and s* and t* are the optimal thresholds.

[0102] 4) Segment the image using the optimal threshold:

[0103]

[0104] Among them, I(x, y) is the gray level at the pixel (x, y).

[0105] S3.2. Calculate the surface roughness of the segmented region:

[0106] Adopt the pixel tracking edge detection method to obtain the edge curve of the segmented region, and perform median line fitting on the edge curve data by the least square method. Combine the edge curve and the median line to obtain the height feature;

[0107] S3.3. Corrosion defect determination: Use the height feature to quantitatively represent the surface roughness of the fan blade. When the height feature value is greater than a certain threshold, it is determined as a corrosion defect.

[0108] S3.2 includes the following steps:

[0109] S3.2.1: Define the upper left corner coordinates of the segmented area as the origin of coordinates, specify the horizontal direction as the x-axis, and the vertical direction as the y-axis to construct a rectangular coordinate system as follows:

[0110] 1) Determine the bounding box of the segmented area:

[0111]

[0112] Where B(x, y) is the binary image matrix after OTSU segmentation, x and y represent the horizontal and vertical positions of the pixel points respectively, x min , x max represent the maximum and minimum values of the horizontal position respectively, and y min , y max represent the maximum and minimum values of the vertical position respectively. Width and Height represent the width and height of the bounding box.

[0113] 2) Define the local coordinate axes:

[0114] The coordinates of the origin of coordinates (x0, y0) are (x min , y min ), the direction of the x-axis is horizontally to the right, and the direction of the y-axis is vertically downward.

[0115] S3.2.2: Adopt the pixel tracking edge detection method, combine the information of the neighboring area of the pixel point, and select the most suitable point in the neighboring area as the next boundary point to obtain the expression of the edge curve. The drawing method of the edge curve is the same as that in step 2.1.

[0116] S3.2.4: Use the least squares method to fit the midline of the edge curve, and its formula is as follows:

[0117] g(x i ) = mx i + n (32);

[0118]

[0119] Where G represents the sum of the squares of the errors between the actual curve and the fitted midline; n' represents the total number of pixel points; g(x i ) and f(x i ) represent the edge curve equation and the midline equation respectively; x i represents the abscissa of each pixel point on the edge curve, and i represents the number of the edge pixel points; m and n represent the slope and intercept of the fitted midline respectively, and the values of m and n that minimize the sum of squares value G need to be calculated.

[0120] In S3.3, the corrosion defect judgment formula is as follows:

[0121]

[0122] Among them, H represents the height eigenvalue; for the blades in actual applications, the surface cannot be made absolutely smooth, so a surface roughness threshold R needs to be specified. a , when the height feature is greater than R a , it is determined as a corrosion defect. R a takes 0.4 μm.

[0123] A method for detecting defects of offshore wind turbine blades based on image enhancement and Hough transform according to the present invention has the following beneficial effects: 1) The present invention uses an image enhancement strategy of histogram equalization and gamma correction for the collected images, which can enhance the image contrast, suppress image overexposure, and at the same time solve the problem of excessive local enhancement caused by the CLAHE algorithm.

[0124] 2) The present invention uses an improved Hough transform and a pixel tracking edge detection method to identify the wind turbine blades. The pixel point domain is divided into eight adjacent regions and pixel filling operations are performed, which can reduce the interference of other objects with linear features such as clouds and improve the detection accuracy of the wind turbine blades;

[0125] 3) The present invention uses OTSU binary segmentation to divide the image into foreground and background parts, obtains the edge curve and the center line of the segmented region to obtain the height feature, and then uses the height feature to quantitatively represent the roughness of the wind turbine blade surface, which can obtain the corrosion condition of the wind turbine blade surface without contact and avoid causing additional damage to the object to be measured. Description of the Drawings

[0126] The present invention will be further described below with reference to the drawings and embodiments:

[0127] Figure 1 is the flow chart of the corrosion detection of the offshore wind turbine blade in the embodiment of the present invention.

[0128] Figure 2 is the pre-image enhancement figure in the embodiment of the present invention.

[0129] Figure 3 is the schematic diagram of the selection of local anchor points in the embodiment of the present invention.

[0130] Figure 4 is the schematic diagram of edge pixel detection in the embodiment of the present invention.

[0131] Figure 5 is the effect diagram of the improved Hough transform for positioning the wind turbine blade in the embodiment of the present invention.

[0132] Figure 6 is the effect diagram of the defect detection of the wind turbine blade in the embodiment of the present invention. Specific Embodiments

[0133] A method for detecting defects of offshore wind turbine blades based on image enhancement and Hough transform, comprising: taking images of offshore wind turbine blades by an unmanned aerial vehicle; enhancing the captured images to reduce overexposure caused by light; constructing a complete contour of the images and positioning the wind turbine blades; segmenting the images and extracting the edge curves and centerlines of the segmented regions; calculating the height features of the positioning regions using the edge curves and centerlines, and characterizing the corrosion condition of the regions through the height features. The present invention improves the detection algorithm, improves the accuracy of wind turbine blade positioning, combines the concept of roughness with corrosion detection, and detects corrosion defects by calculating the surface roughness of the wind turbine blades, thereby improving the detection accuracy of small corrosion.

[0134] As Figure 1 shown, a method for detecting defects of offshore wind turbine blades based on image enhancement and improved Hough transform comprises the following steps:

[0135] Step 1. Image enhancement of offshore wind turbine blades:

[0136] Step 1.1: Convert the image from the RGB space to the HSV (Hue, Saturation, Value) space and extract the V channel;

[0137] Step 1.2: Adjust the contrast and exposure of the image using the CLAHE algorithm in the V channel;

[0138] Step 1.3: Correct the probability density function (pdf) and cumulative distribution function (cdf) of the image and adjust the local mean coefficient K to improve the detail correction effect of the local scene. The formula is as follows:

[0139]

[0140] where pdf n (l) represents the corrected probability density function; pdf(l) represents the current image gray level distribution; pdf max and pdf min represent the set upper and lower limits of the probability distribution; pdf cow represents the compensation component; ε represents the proportional reduction factor, 0 < ε < 1.

[0141]

[0142] where cdf Ω (l) represents the cumulative distribution function value of the gray level l in the corrected window region.

[0143]

[0144]

[0145] Among them, c is a constant, where 0 < c < 1; represents the average luminance within the current local window region; Ω represents the local window; l represents the luminance value of the central pixel of the local window after histogram equalization processing; l max represents the maximum value of the image gray level; O(l) represents the output luminance value after local adaptive gamma correction. Figure 2 is the comparison graph after image enhancement.

[0146] Step 2: Blower blade positioning based on the Hough transform:

[0147] Step 2.1: Object detection based on the pixel tracking edge extraction method:

[0148] First, select a local anchor point as the starting point for detection, and then connect each edge pixel one by one along the image edge direction, finally constructing a series of smooth curves. The selection method of the local anchor point is as Figure 3 shown. The numbers marked in the brackets indicate the inner product energy values of the pixel points. The solid arrows are responsible for indicating the gradient direction of the pixels, and the dashed arrows are responsible for indicating the edge direction of the pixels. In Figure 3 subfigure a, the pixel gradient directions in the edge direction of the neighborhood of point X are inconsistent, so it is determined as an unqualified anchor point. In Figure 3 subfigure b, the inner product energy value of point X in the gradient direction is not the highest, and it is also recognized as an unqualified anchor point. Only in Figure 3 subfigure c, point X satisfies both of the above conditions and is confirmed as a qualified anchor point;

[0149] The detection process of the edge pixel points is as Figure 4 shown. First, the edge tracking direction is divided into 8 parts, and the specific path of each part of the edge tracking is determined by the direction encoding value c. Figure 4 Subfigure b of shows the search area for the next edge pixel point. A0 represents the previous edge pixel point, and A1 represents the current edge pixel point. If the given direction encoding value of the current edge pixel point is c, then the search area for the subsequent edge pixel points can be accurately defined according to the rule {c, (c + 1) & 0x7, (c - 1) & 0x7}.

[0150] Step 2.2: Object positioning based on the improved Hough line detection:

[0151] According to the edge point position information, the pixel neighborhood is divided into eight adjacent regions and pixel filling operations are performed. After obtaining the straight-line pixel points, all the pixel point information belonging to the same straight line is fitted by the least square method. The positioning effect diagram is as Figure 5 shown, and it can be seen from Figure 5 that through the eight-neighborhood filling and least square fitting of the method of the present invention, only the accurate straight-line positioning of the fan blade is retained;

[0152] Step 3: Identification of blade surface defects based on OTSU:

[0153] Step 3.1: Binary segmentation based on OTSU divides the image into two parts, foreground and background;

[0154] Step 3.2: Calculate the surface roughness of the segmented region:

[0155] The pixel tracking edge detection method is used to obtain the edge curve of the segmented region, and the least square method is used to fit the median line of the edge curve data. The height feature is obtained by combining the edge curve and the median line;

[0156] Step 3.3: Corrosion defect determination: The height feature is used to quantitatively represent the roughness of the fan blade surface. When the height feature value is greater than a certain threshold, it is determined as a corrosion defect. The formula for defect determination is as follows:

[0157]

[0158] where H represents the height feature value. For the blades in practical applications, the surface cannot be absolutely smooth, so a surface roughness threshold R a needs to be specified. When the height feature is greater than R a , it is determined as a corrosion defect. R a takes 0.4 μm. The defect detection effect diagram is as Figure 6 shown, and it can be seen from Figure 6 that in the defect detection with different background interferences, the method of the present invention can detect and locate the corrosion defects.

Claims

1. A method for detecting defects in offshore wind turbine blades based on image enhancement and Hough transform, characterized in that It includes the following steps: Step 1: Enhance the images of the offshore wind turbine blades captured by the drone; Step 2: Locate the wind turbine blades based on the Hough transform; Step 3: Identify the surface defects of the wind turbine blades based on OTSU.

2. The method for detecting defects of an offshore wind turbine blade based on image enhancement and Hough transform according to claim 1, wherein: The said Step 1 includes the following steps: S1.1: Convert the image of the wind turbine blade from the RGB space to the HSV space and extract the V channel; S1.2: Adjust the contrast and exposure of the image using the CLAHE algorithm within the V channel; S1.3: Correct the probability density function and cumulative distribution function of the image and adjust the local mean coefficient K to improve the detail correction effect of the local scene.

3. The method for detecting defects of an offshore wind turbine blade based on image enhancement and Hough transform according to claim 2, wherein: In S1.1, convert the image of the wind turbine blade from the RGB space to the HSV (Hue, Saturation, Value) space and extract the V channel; specifically as follows: 1): Map the R, G, B components of each pixel of the image to the interval (0, 1), and the formula is as follows: where R, G, B respectively represent the values of the original pixel channels of the image, 255 is the value range of each channel pixel value, and R', G', B' respectively represent the pixel values of each channel after normalization, and their range is (0, 1); 2): Directly extract the V channel. The V channel represents the brightness information of the image, and its value is the maximum value among R', G', B'. The formula is as follows: V = max(R', G', B') (2).

4. The method for detecting defects of an offshore wind turbine blade based on image enhancement and Hough transform according to claim 2, characterized in that: In S1.2, adjust the contrast and exposure of the image using the CLAHE algorithm within the V channel; specifically as follows: 1) Divide the V channel image into multiple 8×8 sub-blocks and perform mirror padding on the image edges. The formula is as follows: where V(x, y) represents the V channel value of the image at the pixel point (x, y), and V'(x, y) represents the V channel value of the padded image, and Width and Height respectively represent the width and height of the original image; 2) Calculate the local histogram and limit the contrast: where k is the gray level, L is the total number of gray levels, α is the contrast limiting factor, h(k) represents the original frequency of gray level k within the sub-block, and h clip (k) is the frequency of gray level k after clipping, and h final is the final frequency after compensation; 3) Histogram equalization: T(k) is the gray-scale mapping function, L is the total number of gray levels, N is the total number of pixels within the sub-block, and h final is the final frequency after compensation; 4) Bilinear interpolation to merge sub-blocks: Among them, 4 is the number of adjacent sub-blocks of the current pixel, including the upper left, upper right, lower left, and lower right; γ i is the weight, determined by the distance from the pixel to the sub-block; T i is the gray-scale mapping function of the i-th sub-block; V’(x,y) represents the V-channel value of the filled image, and V out (x,y) is the finally output V-channel value.

5. The method for detecting defects of an offshore wind turbine blade based on image enhancement and Hough transform according to claim 2, wherein: S1.3 includes the following steps: 1): Based on the statistical distribution of the image gray levels, use the average strategy to achieve the purpose of histogram equalization, thereby correcting the probability density function of the image. The formula is as follows: Among them, pdf n (l) represents the corrected probability density function; pdf(l) represents the current image gray level distribution; pdf max and pdf min represent the set upper and lower limits of the probability distribution; pdf cow represents the compensation component; ε represents the scaling factor, 0 < ε < 1; 2): Correct the cumulative distribution function of the image. The formula is as follows: Among them, cdf Ω (l) represents the cumulative distribution function value of the gray level l in the corrected window area; 3): Set a local mean coefficient K to adjust the gamma coefficient. This value determines whether the gamma value is greater than 1. Finally, obtain the output O(l) after local adaptive gamma correction. The formula is as follows: Among them, c is a constant, where 0 < c < 1; represents the average luminance within the current local window region; Ω represents the local window; l represents the luminance value of the center pixel of the local window after histogram equalization processing; l max represents the maximum value of the image gray level; O(l) represents the output luminance value after local adaptive gamma correction.

6. The method for detecting defects of an offshore wind turbine blade based on image enhancement and Hough transform according to claim 1, wherein: The said Step 2 includes the following steps: S2.1: Object detection based on the pixel tracking edge extraction method; S2.2: Object location based on the improved Hough line detection; According to the edge point position information, divide the pixel point neighborhood into eight adjacent regions and perform pixel filling operations. After obtaining the line pixel points, fit all the pixel point information belonging to the same line by the least squares method.

7. The method for detecting defects of an offshore wind turbine blade based on image enhancement and Hough transform according to claim 6, wherein: In S2.1, first, select the local anchor point as the starting point for detection. A qualified local anchor point needs to meet the following two conditions simultaneously: 1) Have the same gradient direction as the adjacent pixels in the edge direction; 2) The inner product energy value is the largest among adjacent pixels in the pixel gradient direction; Subsequently, connect each edge pixel one by one along the image edge direction: The initial anchor point connects the discrete edge points from the left and right directions using the edge expansion direction; the tracking direction of subsequent edge pixel points is determined by the edge direction of the edge pixel points; the tracking direction of subsequent edge pixel points is determined by the edge direction of the edge pixel points; Edge pixel detection needs to meet the following conditions: 1) The amplitude of the inner product energy of the edge point is the largest; 2) The edge point is a maximum point within its neighborhood; 3) The edge direction of the edge point is the same as that of the previous edge point; Finally, construct a series of smooth curves: The candidate pixel points that meet the specific three criteria will be evaluated for confidence according to the preset weight; the criteria for confidence evaluation are as follows: 1) Calculate the gradient amplitude G(x,y): Among them, G x represents the gradient in the horizontal direction, and G y represents the gradient in the vertical direction; 2) Direction consistency D(c,c’): D(c,c') = cos(θ(c) - θ(c')) (12); where c is the current direction code, c’ is the candidate direction code, and θ(c) and θ(c’) are the angles corresponding to the direction codes respectively; 3) Neighborhood continuity N(x,y): where the neighborhood window size is 3×3; 4) Calculate the confidence Confidence(x,y) by combining the above three criteria: where ω1, ω2, ω3 are weights satisfying ω1 + ω2 + ω3 = 1, and G max is the maximum gradient magnitude of the image; Determine the pixel point with the highest confidence as the final edge pixel point, and make corresponding adjustments to the edge tracking direction; if no continuous edge pixel points meeting the conditions can be detected in a specific direction, it can be inferred that the edge tracking process in this direction has reached the termination condition; at this time, the tracking process will turn to another direction to continue the edge detection; by fusing the sets of edge pixel points obtained in two independent directions, the complete edge contour of the image is constructed.

8. The method for detecting defects of an offshore wind turbine blade based on image enhancement and Hough transform according to claim 6, characterized in that: S2.2 includes the following steps: S2.2.1: Divide the pixel neighborhood into eight adjacent regions according to the edge point position information and perform pixel filling operations; S2.2.2: Calculate the distance from the pixel points belonging to the straight line to the straight line and set a threshold. The specific steps are as follows: 1) The straight line equation detected by the Hough transform is: y = ax + b (15); where a and b represent the slope and intercept of the straight line respectively; 2) Calculate the distance d from the pixel point to the straight line i : where x i and y i represent the horizontal and vertical coordinates of the pixel point respectively, and a and b represent the slope and intercept of the straight line respectively as above; 3) Set the distance threshold T: T = k·σ (17); where k is the adjustment coefficient and σ is the standard deviation of the noise, and the calculation formula is as follows: Among them, the pixel values extracted from the solid color area are denoted as the set {v1, v2, …, v n}, and μ is the mean value of the pixels in this area; S2.2.3: Determine the distance value d i Compare it with the threshold value T. If the distance value is less than the threshold, it is determined as a linear point pixel and included in the least squares fitting to improve the fitting accuracy. The fitting process is as follows: Among them, (x i , y i ) represents the position of the straight-line pixel points, where x i and y i respectively represent the horizontal and vertical coordinates of the pixel points, and a and b represent the straight-line parameters to be fitted; and represent the optimal parameter estimates; W represents the error function of the straight-line fitting, and N is the number of pixel points.

9. The method for detecting defects of an offshore wind turbine blade based on image enhancement and Hough transform according to claim 1, wherein: Step 3 includes the following steps: S3.1: Divide the image into foreground and background parts based on OTSU-based binary segmentation; specifically as follows: 1) Construct a two-dimensional histogram, and the formula is as follows: where i and j respectively represent the gray level of the pixel point (x, y) and the average gray value within the 3×3 neighborhood centered on the pixel point (x, y), and P ij is the probability that the number of pixels with gray level i and neighborhood mean j appears; 2) Define the weight and mean: Let the probabilities of the background and foreground appearance be λ1 and λ2 respectively, and the corresponding mean vectors be μ1 and μ2 respectively. The mean vector corresponding to the entire image is μ, s and t are the thresholds within the ranges of i and j respectively, and the relevant calculation formulas are as follows: μ=(∑iP ij ,∑jP ij )(27); 3) Calculate the between-class variance trace and select the optimal threshold, and the formula is as follows: tr(S b ) = λ1[(μ 1i - μ i ) 2 + (μ 1j - μ j ) 2 + λ2[(μ 2i - μ i ) 2 + (μ 2j - μ j ) 2 (28); Among them, S b is a discrete measure matrix, and s* and t* are the optimal thresholds; 4) Segment the image using the optimal threshold: where I(x,y) is the gray value at the pixel (x,y); S3.2: Calculate the surface roughness of the segmented region The edge curve of the segmentation region is obtained by using the pixel tracking edge detection method, and the midline fitting is performed on the edge curve data by the least squares method. The height feature is obtained by combining the edge curve and the midline; S3.

3. Corrosion defect determination: The surface roughness of the fan blade is quantitatively represented by the height feature. When the height feature value is greater than a certain threshold, it is determined as a corrosion defect; The corrosion defect determination formula is as follows: Among them, H represents the height eigenvalue; a surface roughness threshold R is specified a , when the height eigenvalue is greater than R a , it is determined as a corrosion defect.

10. The method for detecting defects of an offshore wind turbine blade based on image enhancement and Hough transform according to claim 9, characterized in that: S3.2 includes the following steps: S3.2.1: Define the upper left corner coordinates of the segmentation region as the coordinate origin, specify the horizontal direction as the x-axis, and the vertical direction as the y-axis to construct a rectangular coordinate system as follows: 1) Determine the bounding box of the segmentation region: Among them, B(x, y) is the binary image matrix after OTSU segmentation, where x and y respectively represent the horizontal and vertical positions of the pixel points, and x min , x max represent the maximum and minimum values of the horizontal position respectively, and y min , y max represent the maximum and minimum values of the vertical position respectively; Width and Height represent the width and height of the bounding box respectively; 2) Define the local coordinate axes: The coordinates of the origin of coordinates (x0, y0) are (x min , y min ), the direction of the x-axis is horizontally to the right, and the direction of the y-axis is vertically downward; S3.2.2: Adopt the pixel tracking edge detection method, combine the information of the adjacent region of the pixel point, and select the most suitable point in the adjacent region as the next boundary point to obtain the expression of the edge curve; S3.2.4: Perform midline fitting on the edge curve by the least squares method, and its formula is as follows: g(x i ) = mx i + n (32); Among them, G represents the sum of squared errors between the actual curve and the fitted median line; n' represents the total number of pixel points; g(x i ) and f(x i ) represent the edge curve equation and the median line equation respectively; x i represents the abscissa of each pixel point on the edge curve, and i represents the number of edge pixels; m and n represent the slope and intercept of the fitted median line respectively, and the values of m and n that minimize the sum of squared values G need to be calculated.

Images