Visual identification method and system for abrasion of composite coating of water turbine

Abstract

The invention belongs to the technical field of water turbine detection based on image recognition, and discloses a visual recognition method and system for abrasion of a composite coating of a water turbine, compared with traditional manual detection, the efficiency is improved, abrasion recognition does not depend on professional experience of personnel, and therefore the detection accuracy is improved; meanwhile, in the detection process, the actual working environment of the water turbine blade is considered, and deblurring, color correction, contrast and detail enhancement and other processing are performed on the blade image specifically, so that the interference of the underwater complex environment on blade coating abrasion detection can be reduced, the limitation of coating abrasion automatic identification application is solved, and the detection accuracy is improved. Furthermore, the interference of noise on the abrasion identification of the coating can be reduced, so that the identification accuracy is further improved; therefore, the invention provides a coating abrasion automatic detection technology with high efficiency and accuracy, and the technical blank in the field of water turbine blade coating automatic abrasion detection is filled.

Description

Technical Field

[0001] This invention belongs to the field of image recognition-based turbine inspection technology, specifically relating to a method and system for visually identifying wear on composite coatings of turbines. Background Technology

[0002] Turbine blades face significant water flow impact and corrosion during operation. Therefore, applying anti-erosion coatings to the blades is an effective measure to improve the service life of turbines and ensure the safe operation of hydropower stations. Currently, various types of anti-erosion coating materials for turbines have been developed, such as metal-based coatings, ceramic-based coatings, and polymer composite coatings. These coating materials have excellent anti-erosion properties, corrosion resistance, and adhesion, effectively protecting turbine blades from wear and corrosion, thereby extending the service life of the turbine.

[0003] However, during long-term operation, the coating on turbine blades is prone to peeling and cracking, which in severe cases can lead to deformation and failure of the blade substrate, resulting in the scrapping of a large number of blades. Therefore, wear detection of composite coatings on turbine blades is an important measure to ensure the safe operation of turbines. Currently, coating wear assessment is mainly carried out through manual visual inspection or simple image comparison, which is not only inefficient, but also highly subjective and inaccurate due to the dependence of wear detection on the professional skills of the personnel. At the same time, the blades are underwater for a long time, and the complex underwater environment also brings significant interference to the wear detection of turbine blades, thus limiting the application of automated coating wear identification. Therefore, based on the aforementioned shortcomings, there is an urgent need to provide a highly efficient and accurate visual identification method for wear of composite coatings on turbines to fill the technological gap in the field of automated wear detection of turbine blade coatings. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for visually identifying wear of composite coatings in water turbines, in order to solve the problems of low efficiency, low accuracy, and susceptibility to interference from complex underwater environments in existing technologies.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, a method for visually identifying wear on composite coatings of hydraulic turbines is provided, including: Acquire images of the turbine blades and determine the calibration color information of the blade images; Based on the calibrated color information, the leaf image is deblurred and color corrected to obtain the corrected image; The corrected image is subjected to contrast enhancement processing and detail enhancement processing respectively to obtain a contrast-enhanced image and a detail-enhanced image. The contrast-enhanced image and the detail-enhanced image are then fused to obtain an enhanced leaf image. The enhanced blade image is segmented to obtain the coating wear area; Feature extraction processing is performed on the coating wear area to obtain the texture features and geometric features of the blade image, and the coating wear features are constructed using the texture features and geometric features; The coating wear characteristics are input into the coating wear identification model to obtain the coating wear identification results of the turbine blades.

[0006] Based on the above-disclosed content, after acquiring an image of a turbine blade, this invention first determines the calibration color information of the blade image. Then, based on the calibration color information, it performs deblurring and color correction processing on the blade image to obtain a corrected image. This improves image clarity and balances color, thereby reducing color distortion. Next, the invention performs contrast enhancement and detail enhancement on the corrected image, and then fuses the two enhanced images to obtain an enhanced blade image. This solves the problems of insufficient local contrast and uneven lighting in underwater images, making the image texture more prominent and highlighting image details. After image enhancement, image segmentation is performed, specifically segmenting the coating wear area. Then, texture and geometric features are extracted from the coating wear area to obtain coating wear features. Finally, the coating wear features are input into the coating wear recognition model to obtain the coating wear recognition result of the turbine blade.

[0007] Through the above design, compared with traditional manual inspection, this invention not only improves efficiency, but also enhances the accuracy of wear identification by eliminating reliance on personnel's professional experience. Furthermore, considering the actual working environment of the turbine blades during inspection, targeted processing such as deblurring, color correction, contrast enhancement, and detail enhancement is applied to the blade images. This reduces the interference of the complex underwater environment on blade coating wear detection, thus overcoming the limitations of automated coating wear identification applications. It also reduces noise interference in coating wear identification, further improving accuracy. Therefore, this invention provides a highly efficient and accurate automated coating wear detection technology, filling a technological gap in the field of automated wear detection for turbine blade coatings. Consequently, it is highly suitable for large-scale application and promotion.

[0008] In one possible design, the calibration color information of the blade image is determined, including: Initialize the number of searches i, and obtain the target image at the time of the i-th search, where the initial value of i is 1, and when i is 1, the target image at the time of the i-th search is the leaf image; The target image is divided into equal parts to obtain several image blocks; Calculate the pixel variance of each image patch and select the image patch with the largest pixel variance as the optimal image patch for the i-th search. Determine whether the search stopping condition is met, wherein the search stopping condition is that the total number of pixels in the optimal image block is less than or equal to a pixel count threshold; If not, the optimal image patch in the i-th search will be used as the target image in the (i+1)-th search. Increment i by 1 and reacquire the target image at the time of the i-th search until the search stopping condition is met, so that the optimal image block at the time of the search stopping condition is used as the calibration image block; The calibration color information is determined based on the calibration image block.

[0009] In one possible design, based on the calibrated color information, the leaf image is deblurred and color corrected to obtain a corrected image, including: Based on the calibrated color information, the blur correction factor of the leaf image in different channels is determined; The leaf image is deblurred using the blur correction factors on different channels and the calibrated color information to obtain a deblurred image; Based on the calibrated color information, the color compensation type of the deblurred image is determined; Based on the color compensation type of the deblurred image, determine the color compensation parameters; Based on the color compensation parameters, the deblurred image is subjected to color correction processing to obtain the corrected image.

[0010] In one possible design, the calibration color information includes the calibration colors of the leaf image on three color channels, and the three color channels include the R channel, the G channel, and the B channel; Specifically, based on the calibrated color information, the blur correction factors for the leaf image in different channels are determined, including: Based on the calibrated colors in the three color channels, the channel values ​​of each pixel in the leaf image are normalized in the three color channels to obtain the normalized channel values ​​of each pixel in the three color channels. For any color channel, the smallest normalized channel value is selected from the normalized channel values ​​of each pixel in that color channel and used as the calibration channel value. After all color channels have been queried, three calibration channel values ​​are obtained. The smallest calibration channel value is selected from the three calibration channel values, and the fuzzy correction factor corresponding to the R channel is calculated based on the smallest calibration channel value. The first and second deblurring factors are calculated based on the calibrated colors in the three color channels. The fuzz correction factor for the G channel is determined using the fuzz correction factor and the first defuzzification factor corresponding to the R channel, and the fuzz correction factor for the B channel is determined using the fuzz correction factor and the second defuzzification factor corresponding to the R channel.

[0011] In one possible design, the corrected image undergoes contrast enhancement processing, including: The leaf image and the corrected image are converted to HSV space respectively to obtain the first HSV image and the second HSV image; Calculate the first mean and first standard deviation of the S channel in the first HSV image, and the second mean and second standard deviation of the V channel in the first HSV image; Calculate the third mean and third standard deviation of the S channel in the second HSV image, and the fourth mean and fourth standard deviation of the V channel in the second HSV image; The contrast enhancement factor of the S channel is calculated based on the first mean, the third mean, the first standard deviation, and the third standard deviation; and the contrast enhancement factor of the V channel is calculated based on the second mean, the fourth mean, the second standard deviation, and the fourth standard deviation. By using the contrast enhancement factors of the S channel and the V channel, the contrast of the S channel and V channel in the second HSV image is enhanced to obtain an enhanced HSV image. The enhanced HSV image is converted to RGB space to obtain the contrast-enhanced image.

[0012] In one possible design, the contrast-enhanced image and the detail-enhanced image are fused to obtain an enhanced leaf image, including: The first image fusion weight of each pixel in the contrast-enhanced image is determined, and a first initial feature weight map corresponding to the contrast-enhanced image is generated based on the first image fusion weight of each pixel in the contrast-enhanced image. The second image fusion weight of each pixel in the detail-enhanced image is determined, and a second initial feature weight map corresponding to the detail-enhanced image is generated based on the second image fusion weight of each pixel in the detail-enhanced image. The first initial feature weight map and the second initial feature weight map are regularized to obtain the first feature weight map and the second feature weight map. Using the first feature weight map and the second feature weight map, the contrast-enhanced image and the detail-enhanced image are fused to obtain the enhanced leaf image after image fusion.

[0013] In one possible design, the first image fusion weight for each pixel in the contrast-enhanced image is determined, including: The contrast-enhanced image is converted into a Lab image, and the channel mean of the three channels in the Lab image is calculated. Based on the channel mean of the three channels and the Lab value of each pixel in the Lab image, the contrast weight of each pixel in the contrast-enhanced image is calculated. Based on the RGB values ​​of each pixel in the contrast-enhanced image, the color weight and exposure weight of each pixel in the contrast-enhanced image are calculated. The first image fusion weight of each pixel is determined by using the contrast weight, color weight, and exposure weight of each pixel in the contrast-enhanced image.

[0014] In one possible design, image fusion is performed on the contrast-enhanced image and the detail-enhanced image using the first feature weight map and the second feature weight map, including: A first multi-level pyramid is constructed for the contrast-enhanced image and the detail-enhanced image, wherein the first multi-level pyramid includes multiple layers of images, and each layer of images has a different resolution; Construct the second multi-level pyramid corresponding to the first feature weight map and the second feature weight map respectively; By fusing the images of the two first-level multi-level pyramids and the two second-level multi-level pyramids at each level, a multi-level fused image pyramid is obtained. The enhanced leaf image is obtained by upsampling the image from the top layer of the fused image pyramid until the bottom layer of the fused image pyramid has been upsampled.

[0015] In one possible design, the enhanced blade image is subjected to image segmentation processing to obtain the coating wear area, including: Edge detection processing is performed on the enhanced blade image to obtain an edge-detected image; Morphological operations are performed on the edge detection image to obtain the coating wear area after the morphological operations; Correspondingly, feature extraction processing for the worn areas of the coating includes: Calculate the gray-level co-occurrence matrix of the wear area of ​​the coating; Based on the gray-level co-occurrence matrix, several texture statistics are determined; Calculate the LBP histogram of the wear area of ​​the coating; The texture features are generated using several texture statistics and the LBP histogram; Extract the centroid coordinates, area, relative area, circularity, rectangularity, and triangularity of the wear area of ​​the coating; The geometric features are constructed using centroid coordinates, area, relative area, circularity, rectangularity, and triangularity.

[0016] Secondly, a visual identification system for wear of composite coatings on water turbines is provided, including: The calibration unit is used to acquire images of the turbine blades and determine the calibration color information of the blade images. The image correction unit is used to perform deblurring and color correction processing on the leaf image according to the calibration color information to obtain the corrected image; An image enhancement unit is used to perform contrast enhancement processing and detail enhancement processing on the corrected image respectively to obtain a contrast-enhanced image and a detail-enhanced image, and to perform image fusion on the contrast-enhanced image and the detail-enhanced image to obtain an enhanced leaf image; An image segmentation unit is used to perform image segmentation processing on the enhanced blade image to obtain the coating wear area; The feature extraction unit is used to perform feature extraction processing on the coating wear area to obtain the texture features and geometric features of the blade image, and to construct the coating wear features using the texture features and the geometric features. The wear identification unit is used to input the coating wear characteristics into the coating wear identification model to obtain the coating wear identification result of the turbine blade.

[0017] Thirdly, a device for visually identifying wear of composite coatings on water turbines is provided. Taking the device as an electronic device as an example, it includes a memory, a processor, and a transceiver that are connected in sequence. The memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the method for visually identifying wear of composite coatings on water turbines as described in the first aspect or any possible design of the first aspect.

[0018] Fourthly, a storage medium is provided, on which instructions are stored, which, when executed on a computer, perform the turbine composite coating wear visualization and identification method as described in the first aspect or any possible design of the first aspect.

[0019] Fifthly, a computer program product containing instructions is provided, which, when executed on a computer, causes the computer to perform the turbine composite coating wear visualization and identification method as described in the first aspect or any possible design of the first aspect.

[0020] Beneficial effects: Compared to traditional manual inspection, this invention not only improves efficiency but also enhances the accuracy of wear detection by eliminating reliance on personnel's professional experience. Furthermore, considering the actual working environment of turbine blades during inspection, targeted processing such as deblurring, color correction, contrast enhancement, and detail enhancement is applied to the blade images. This reduces interference from the complex underwater environment on blade coating wear detection, overcoming limitations in automated coating wear detection applications. It also reduces noise interference, further improving accuracy. Therefore, this invention provides a highly efficient and accurate automated coating wear detection technology, filling a technological gap in the field of automated wear detection for turbine blade coatings. Consequently, it is highly suitable for large-scale application and promotion. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the steps of the method for visually identifying wear of composite coatings on water turbines provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of the water turbine composite coating wear visualization and recognition system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0023] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0024] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0025] Example: See Figure 1 As shown, the turbine composite coating wear visualization recognition method provided in this embodiment, after acquiring the turbine blade image, first determines the calibration color information of the blade image. Then, based on the calibration color information, the blade image is deblurred and color corrected to obtain a corrected image. This improves image clarity and balances image color, thereby improving color distortion. Subsequently, the present invention performs contrast enhancement and detail enhancement on the corrected image, and then fuses the two enhanced images to obtain an enhanced blade image. Based on this, the problems of insufficient local contrast and uneven illumination in underwater images can be solved, and the image texture is made more prominent, highlighting more image details. After image enhancement, image segmentation can be performed, i.e., segmentation... The coating wear area is identified; then, texture and geometric features are extracted from the coating wear area to obtain coating wear features; finally, the coating wear features are input into the coating wear recognition model to obtain the coating wear recognition result of the turbine blade. Thus, this method provides a highly efficient and accurate automatic coating wear detection technology, filling the technical gap in the field of automatic coating wear detection for turbine blades. Therefore, it is very suitable for large-scale application and promotion. For example, this method can be run on the coating wear detection end, and optionally, the coating wear detection end can be, but is not limited to, a server and a computer. It is understood that the aforementioned execution subject does not constitute a limitation on the embodiments of this application. Accordingly, the operation steps of this method can be, but are not limited to, as shown in steps S1 to S6 below.

[0026] S1. Acquire images of the turbine blades and determine the calibration color information of the blade images. In specific implementations, for example, but not limited to, using an underwater camera to capture images of the turbine blades, the corresponding calibration color information can be determined after acquiring the blade images, so that the blade images can be deblurred and color corrected based on the calibration color information. In this embodiment, a calibration color search method based on recursive image segmentation is provided, and the process can be, but is not limited to, the steps S11 to S17 below.

[0027] S11. Initialize the number of searches i and obtain the target image at the time of the i-th search, where the initial value of i is 1, and when i is 1, the target image at the time of the i-th search is the leaf image; in this embodiment, when performing the first search, the leaf image obtained in step S1 can be used as the target image, and then it is divided into equal parts to obtain multiple image blocks of the same size, as shown in step S12 below.

[0028] S12. Divide the target image into equal parts to obtain several image blocks; in specific implementation, for example, but not limited to, dividing the target image into four equal parts to obtain four image blocks; then, the search and judgment can be performed based on the pixel variance of the four image blocks, as shown in steps S13 and S14 below.

[0029] S13. Calculate the pixel variance of each image block and select the image block with the largest pixel variance as the optimal image block for the i-th search. In specific applications, the following formula can be used, but is not limited to, to calculate the pixel variance of any image block.

[0030] ; In the formula, Let represent the pixel variance of any image patch, where n is the total number of pixels in that image patch. This represents the channel value of the x-th pixel in any image patch on the h channel (where h is R, which is the red channel value; h is G, which is the green channel value; and h is B, which is the blue channel value). These represent the blue, green, and red channel values ​​of the x-th pixel, respectively. This represents the average channel value of any image patch on the h channel, that is, the average channel value on the blue, red, and green channels.

[0031] Thus, as can be seen from the aforementioned formula, the pixel variance calculation fully considers factors such as the pixel mean of each channel (R, G, and B channels), the variance of pixel values, and the differences between channels. Therefore, an image patch with a larger pixel variance indicates that the image is more uniform, has higher brightness, and is less affected by the target object. Based on this, the image patch with the largest pixel variance among the four image patches is selected as the optimal image patch currently searched. Then, the search stopping condition can be determined, as shown in step S14 below.

[0032] S14. Determine whether the search stopping condition is met, wherein the search stopping condition is that the total number of pixels of the optimal image block is less than or equal to the pixel number threshold; in specific implementation, the pixel number threshold may be, but is not limited to, 1% of the total number of pixels of the leaf image; thus, when the total number of pixels of the optimal image block is greater than the aforementioned pixel number threshold, the search needs to continue based on the currently searched optimal image block, and the process is as shown in steps S15 and S16 below.

[0033] S15. If not, the optimal image block in the i-th search is used as the target image in the (i+1)-th search. In this embodiment, the currently searched optimal image block is used as the target image in the next search. Then, the aforementioned steps can be repeated, that is, the image is divided into four equal parts, and the pixel variance is calculated again until the total number of pixels of the optimal image block obtained by the search is less than or equal to the pixel number threshold, then the calibration image block can be obtained.

[0034] The recursive search process is shown in step S16 below.

[0035] S16. Increment i by 1 and reacquire the target image at the time of the i-th search until the search stopping condition is met. The optimal image block that meets the search stopping condition is taken as the calibration image block. After the calibration image block is found based on the above steps, the calibration color information can be determined based on it. The process is shown in step S17 below.

[0036] S17. Determine the calibration color information based on the calibration image block; in this embodiment, for example, but not limited to, the average values ​​of the R, G, and B channels of the calibration image block (i.e., the average values ​​of the R, G, and B channels of each pixel in the calibration image block) can be used as the calibration colors on the three color channels, that is, as the calibration colors corresponding to the R, G, and B channels; then, the calibration color information is composed using the calibration colors corresponding to the aforementioned three channels.

[0037] Thus, through the aforementioned steps S11 to S17, the recursive image segmentation search algorithm can be used to search for the most uniform, brightest, and least affected area in the leaf image. Based on this area, the calibration color can be determined. Then, based on the calibration color, the leaf image can be deblurred and color corrected, as shown in step S2 below.

[0038] S2. Based on the calibrated color information, the leaf image is deblurred and color corrected to obtain a corrected image. In this embodiment, the complexity of the underwater environment and lighting conditions can cause the underwater leaf image to be blurry and have color deviation. If not processed, it will cause adverse interference to the subsequent coating wear identification. Therefore, it is necessary to first perform deblurring and color correction on the image to improve the image clarity and solve the color distortion problem.

[0039] For example, steps S21 to S25 can be used to perform deblurring and color correction.

[0040] S21. Based on the calibrated color information, determine the blur correction factor of the leaf image in different channels; in specific applications, this embodiment calculates the blur correction factor of the leaf image in three color channels to perform deblurring processing on the leaf image in the red, blue and green channels; for example, but not limited to, the following steps S21a to S21e can be used to calculate the blur correction factor of the leaf image in the three channels.

[0041] S21a. Based on the calibrated colors in the three color channels, normalize the channel values ​​of each pixel in the leaf image in the three color channels to obtain the normalized channel values ​​of each pixel in the three color channels. In this embodiment, for the red channel, the channel value of each pixel in the leaf image in the red channel is divided by the calibrated color corresponding to the R channel to obtain the normalized channel value of each pixel in the red channel. Of course, the normalization process for the green and blue channels is the same, and will not be described in detail here.

[0042] After the normalization process of the three color channels is completed, the normalized channel values ​​can be selected, as shown in step S21b below.

[0043] S21b. For any color channel, select the smallest normalized channel value from the normalized channel values ​​of each pixel in that color channel as the calibration channel value. After polling all color channels, three calibration channel values ​​are obtained. In this embodiment, this is equivalent to selecting the smallest normalized channel value in the red channel, the smallest normalized channel value in the green channel, and the smallest normalized channel value in the blue channel as the calibration channel values. Then, select the smallest one from the three calibration channel values ​​to calculate the reference blur correction factor (i.e., the blur correction factor corresponding to the R channel).

[0044] The calculation process of the reference fuzzy correction factor is shown in step S21c below.

[0045] S21c. Select the smallest calibration channel value from the three calibration channel values, and calculate the fuzz correction factor corresponding to the R channel based on the smallest calibration channel value. In specific implementation, the difference between 1 and the smallest calibration channel value is calculated, and the difference is used as the reference fuzz correction factor, which is the fuzz correction factor corresponding to the R channel. Then, the fuzz correction factors of the other two channels can be calculated based on the reference fuzz correction factor, as shown in steps S21d and S21e below.

[0046] S21d. Calculate the first deblurring factor and the second deblurring factor based on the calibrated colors on the three color channels; in this embodiment, the first deblurring factor and the second deblurring factor may be calculated according to, but are not limited to, the following formula.

[0047] ; ; In the formula, These represent the first deblurring factor and the second deblurring factor, respectively. This indicates the calibration color of the R channel. They represent the defuzzification coefficients, This indicates the wavelength of the G channel (value is 540nm). Indicates the calibration color of the G channel. This indicates the wavelength of the R channel (value is 620nm). This indicates the calibration color of channel B. This represents the wavelength of channel B (valued at 450nm); where, The values ​​are 0.00113 and 1.62517, respectively.

[0048] Thus, based on the aforementioned formula, after calculating the first deblurring factor and the second deblurring factor, the fuzzing correction factors corresponding to the G channel and the B channel can be calculated by combining the aforementioned reference fuzzing correction factor, as shown in step S21e below.

[0049] S21e. Using the fuzz correction factor corresponding to the R channel and the first defuzzification factor, determine the fuzz correction factor corresponding to the G channel; and using the fuzz correction factor corresponding to the R channel and the second defuzzification factor, determine the fuzz correction factor corresponding to the B channel; in this embodiment, the fuzz correction factor corresponding to the G channel is: In the formula, These represent the fuzzy correction factors for the G channel and the R channel, respectively; similarly, the fuzzy correction factor for the B channel is: ,in, This represents the fuzz correction factor corresponding to channel B.

[0050] Thus, after calculating the blur correction factor of the leaf image in the three color channels through the aforementioned steps S21a to S21e, the deblurring process can be performed based on this, as shown in step S22 below.

[0051] S22. Using the blur correction factors on different channels and the calibration color information, the leaf image is deblurred to obtain a deblurred image; in specific applications, for example, but not limited to, the following formula can be used to deblur the leaf image.

[0052] ; In the formula, Indicates a deblurred image. This represents the channel value of each pixel in the leaf image in the h channel (i.e., the channel value in the R, B, and G channels). This indicates the calibrated colors corresponding to the three color channels. This represents the blur correction factor corresponding to the three color channels. The minimum blur correction factor (with a value of 0.1, its function is to prevent some areas from being overexposed due to excessive reduction of the modulus caused by a blur correction factor that is too small).

[0053] Based on this, the aforementioned formula performs color correction on each color channel of the leaf image. That is, by substituting the blur correction factor and calibration color of the corresponding color channel, as well as the channel value of each pixel in the corresponding color channel, the correction of the corresponding color channel can be completed. Finally, after traversing all color channels, the deblurred image can be obtained.

[0054] After the leaf image is deblurred, color compensation can be performed. In specific applications, this embodiment provides an adaptive color compensation algorithm based on the calibrated color information. First, the color compensation type of the image is determined according to the calibrated color information. Then, the color compensation parameters are calculated specifically according to different color compensation types, thereby performing targeted color compensation. In this way, the effect of color compensation can be improved.

[0055] Optionally, the process for determining the color compensation type is as shown in step S23 below.

[0056] S23. Based on the calibration color information, determine the color compensation type of the deblurred image; in specific implementation, if the absolute value of the difference between the calibration color of the G channel and the calibration color of the B channel is less than or equal to a preset threshold (set to 0.1), then the color compensation type is determined to be red compensation; similarly, if the difference between the calibration color of the B channel and the calibration color of the G channel is greater than the preset threshold, then the color compensation type is determined to be red-green compensation; furthermore, when the difference between the calibration color of the G channel and the calibration color of the B channel is greater than the preset threshold, and the difference between the calibration color of the R channel and the calibration color of the B channel is less than or equal to the preset threshold, then the color compensation type is determined to be red-blue compensation; finally, if the difference between the calibration color of the G channel and the calibration color of the B channel, and the difference between the calibration color of the R channel and the calibration color of the B channel are both greater than the preset threshold, then the color compensation type is determined to be blue compensation.

[0057] Thus, after determining the color compensation type of the image based on the aforementioned method, the color compensation parameters can be determined according to different compensation types, as shown in step S24 below.

[0058] S24. Determine the color compensation parameters according to the color compensation type of the deblurred image; in this embodiment, if the color compensation type is red compensation, the red compensation parameters are calculated using the following formula; ; In the formula, This represents the red compensation parameter.

[0059] Additionally, if the color compensation type is red-green compensation, then in addition to the red compensation parameter, the green compensation parameter also needs to be calculated; then, the red compensation parameter and the green compensation parameter are used to form the color compensation parameter corresponding to red-green compensation.

[0060] For example, the following formula can be used, but is not limited to, to calculate the green compensation parameter.

[0061] ; In the formula, This represents the green compensation parameter.

[0062] Furthermore, if the color compensation type is red-blue compensation, then the first blue compensation parameter is calculated based on the aforementioned red compensation parameter; then, the red compensation parameter and the first blue compensation parameter can be used to form the color compensation parameter corresponding to red-blue compensation.

[0063] Optionally, for example, but not limited to, the following formula can be used to calculate the first blue compensation parameter.

[0064] ; In the formula, This represents the first blue compensation parameter.

[0065] Finally, if the color compensation type is blue compensation, that is, only the blue channel needs to be compensated, then a second blue compensation parameter that is different from the first blue compensation parameter mentioned above needs to be used.

[0066] Optionally, for example but not limited to, the following formula can be used to calculate the second blue compensation parameter.

[0067] ; In the formula, This represents the second blue compensation parameter.

[0068] Therefore, after calculating the color compensation parameters corresponding to different color compensation types using the aforementioned formulas, color correction can be performed based on these parameters, as shown in step S25 below.

[0069] S25. Based on the color compensation parameters, perform color correction processing on the deblurred image to obtain the corrected image; in specific applications, if the color compensation type is red compensation, then use the red compensation parameters to perform color correction processing on the red channel in the deblurred image, thereby obtaining the corrected image after the correction processing.

[0070] Optionally, the red channel of the deblurred image can be corrected according to the following formula to obtain the corrected image.

[0071] ; In the formula, This represents the red channel image in the corrected image. This represents the red channel image in the deblurred image. This represents the green channel image in the deblurred image.

[0072] Similarly, if the color compensation type is red-green compensation, then in addition to the red channel correction of the deblurred image mentioned above, the green channel correction also needs to be performed.

[0073] For example, the following formula can be used, but is not limited to, to perform green channel correction on a deblurred image; ; In the formula, This represents the green channel image in the corrected image. This represents the blue channel image in the deblurred image; This represents the green channel image in the deblurred image.

[0074] Thus, by correcting the red channel and the green channel, red-green compensation of the deblurred image can be completed, resulting in the corrected image.

[0075] Furthermore, when the color compensation type is red-blue compensation, in addition to the aforementioned red channel correction of the deblurred image, the blue channel also needs to be corrected.

[0076] For example, the following formula can be used, but is not limited to, to correct the blue channel of a deblurred image.

[0077] ; In the formula, This represents the blue channel image in the corrected image.

[0078] Thus, by correcting the red channel and the blue channel, red-blue compensation of the deblurred image can be completed, resulting in the corrected image.

[0079] Finally, when the color compensation type is blue compensation, the aforementioned second blue compensation parameter is used to perform blue channel correction on the deblurred image, thereby obtaining the corrected image.

[0080] Optionally, the blue channel of the deblurred image can be corrected according to the following formula; .

[0081] In the formula, This represents the blue channel image in the corrected image.

[0082] Therefore, after performing deblurring and color correction through the aforementioned steps S21 to S25, the image clarity can be improved and the image color can be balanced, thereby achieving the function of improving color distortion; then, image enhancement can be performed to solve the problems of insufficient local contrast and blurred details in underwater images; the image enhancement process is as shown in step S3 below.

[0083] S3. Perform contrast enhancement processing and detail enhancement processing on the corrected image respectively to obtain a contrast-enhanced image and a detail-enhanced image, and perform image fusion on the contrast-enhanced image and the detail-enhanced image to obtain an enhanced leaf image; In specific applications, this embodiment provides an image enhancement algorithm that calculates the contrast enhancement factor by comparing the original image (i.e., the leaf image) and the corrected image, and performs contrast enhancement based on the contrast enhancement factor.

[0084] The specific process of the aforementioned contrast-enhanced image can be, but is not limited to, the steps S31 to S36 described below.

[0085] S31. Convert the leaf image and the corrected image to HSV space respectively to obtain the first HSV image and the second HSV image; after completing the conversion from RGB space to HSV space, the mean and standard deviation of the S and V channels can be calculated, as shown in steps S32 and S33 below.

[0086] S32. Calculate the first mean and first standard deviation of the S channel in the first HSV image, and the second mean and second standard deviation of the V channel in the first HSV image; in this embodiment, these are the mean and standard deviation of the S channel values ​​of each pixel in the first HSV image, and the mean and standard deviation of the V channel values ​​of each pixel; in the same way as described above, the mean and standard deviation of the S and V channels in the second HSV image can be calculated, as shown in step S33 below.

[0087] S33. Calculate the third mean and third standard deviation of the S channel in the second HSV image, and the fourth mean and fourth standard deviation of the V channel in the second HSV image.

[0088] After calculating the mean and standard deviation of the S and V channels in the two HSV images, the images can be compared to calculate the contrast enhancement factor corresponding to the S and V channels respectively; the calculation process of the contrast enhancement factor is shown in step S34 below.

[0089] S34. Calculate the contrast enhancement factor of the S channel based on the first mean, the third mean, the first standard deviation, and the third standard deviation, and calculate the contrast enhancement factor of the V channel based on the second mean, the fourth mean, the second standard deviation, and the fourth standard deviation; in specific applications, examples, but not limited to, can use the following formula to calculate the contrast enhancement factor of the S channel.

[0090] ; In the formula, This represents the contrast enhancement factor of the S channel. These represent the first and third means, respectively. These represent the first standard deviation and the third standard deviation, respectively. Indicates contrast weight. The value represents the correction parameter. In this embodiment, the contrast weight is 0.2, and the correction parameters are 0.025 and 0.06 respectively. That is, when calculating the contrast factor of the S channel, the value is 0.025, and when calculating the contrast factor of the V channel, the value is 0.06.

[0091] Of course, the calculation process for the contrast enhancement factor of the V channel is the same, and the principle will not be repeated here.

[0092] Thus, after calculating the contrast enhancement factors of the S and V channels, the contrast enhancement of the corrected image can be performed, as shown in step S35 below.

[0093] S35. Using the contrast enhancement factor of the S channel and the contrast enhancement factor of the V channel, the contrast of the S channel and the V channel in the second HSV image is enhanced to obtain an enhanced HSV image; in specific implementation, the process for contrast enhancement of the S channel in the second HSV image is as follows: (1) Extract the S channel image in the second HSV image; (2) Divide the S channel image into multiple non-overlapping local image blocks; (3) Calculate the grayscale histogram of each local image block; (4) Calculate the grayscale cropping threshold (i.e., using the contrast enhancement factor corresponding to the S channel) based on the contrast enhancement factor of the S channel. (3) Multiply the number of pixels in each image block by the contrast enhancement factor to obtain the gray-level cropping threshold, and use the gray-level cropping threshold to crop each gray-level histogram to obtain each cropped gray-level histogram; (4) Perform histogram equalization mapping on each cropped gray-level histogram (such as using the cumulative distribution function CDF for mapping, which is a commonly used function for histogram mapping, and its principle will not be elaborated here) to obtain each enhanced local image block; (5) Stitch together each enhanced local image block to obtain the enhanced S-channel image; In this way, the enhancement of the S-channel can be completed by using the above method.

[0094] In this embodiment, cropping each grayscale histogram means that the portion of the histogram exceeding the cropping threshold is redistributed to all grayscale levels. For example, if a small block has 64 pixels, the contrast enhancement factor is 0.05, and the grayscale cropping threshold is 64 multiplied by 0.05, rounded to the nearest whole number, the grayscale cropping threshold is 3. This means that in this small block, the number of pixels at any grayscale level cannot exceed 3, and the excess will be redistributed to the remaining grayscale levels.

[0095] Similarly, the V channel enhancement process is also the same, and will not be elaborated here.

[0096] After enhancing the contrast of the S and V channels in the second HSV image based on the aforementioned steps, it can be converted to RGB space to obtain a contrast-enhanced image, as shown in step S36 below.

[0097] S36. Convert the enhanced HSV image to RGB space to obtain the contrast-enhanced image.

[0098] Thus, through the aforementioned steps S31 to S36, this embodiment adjusts the contrast between HSV channels based on the comparison between HSV channels. By calculating the mean and standard deviation of the S and V channels of the original image and the corrected image, and using the mean and standard deviation of both as parameters to limit the contrast adaptive histogram equalization, the contrast of the S and V channels of the corrected image is adaptively adjusted. In this way, the contrast of the image can be enhanced without producing layering and noise problems.

[0099] After contrast enhancement is completed, detail enhancement can be performed, and the process is as follows: Step 1: Perform guided filtering on the corrected image to obtain a calibration image. In this embodiment, the corrected image can be set as the guide image. Then, guided filtering is performed on the corrected image based on the guide image to obtain the calibration image, which is the base layer image. Then, the calibration image can be obtained by difference between the calibration image and the corrected image. The process is shown in Step 2 below.

[0100] Step 2: Calculate the difference image between the corrected image and the calibrated image, and perform detail enhancement processing on the difference image to obtain an enhanced difference image. In specific implementation, the largest pixel value can be selected from the difference image first, but is not limited to this step. Then, based on the largest pixel value, the following formula is used to enhance the details of the difference image.

[0101] ; In the formula, Indicates enhanced difference image, Represents a difference image. This represents the maximum pixel value. represents the stretching coefficient, which takes a value of 0.7. sign is the sign function, which extracts the sign of each pixel value in the detail layer image (i.e., the difference image, which contains positive values ​​(representing details brighter than the base layer) and negative values ​​(representing details darker than the base layer)) and restores its original sign after non-linear stretching to maintain the polarity of detail enhancement (i.e., to keep the edges from being bright or dark).

[0102] Therefore, this embodiment introduces a nonlinear stretching function to map the detail layer (i.e., the difference image), thereby enhancing the detail layer.

[0103] Thus, after the detail enhancement of the difference image is completed, it can be fused with the base layer to obtain a detail-enhanced image, as shown in step 3 below.

[0104] Step 3: Perform image fusion between the enhanced differential image and the calibration image to obtain a detail-enhanced image.

[0105] Once the details and contrast of the corrected image are enhanced, image fusion can then be performed.

[0106] This embodiment provides an image fusion algorithm based on feature weights, the specific implementation process of which may be, but is not limited to, the steps S37 to S310 below.

[0107] S37. Determine the first image fusion weight of each pixel in the contrast-enhanced image, and generate the first initial feature weight map corresponding to the contrast-enhanced image based on the first image fusion weight of each pixel in the contrast-enhanced image; In specific implementation, this embodiment generates the first image fusion weight of each pixel by calculating the contrast weight, color weight and exposure weight of each pixel, and the process is shown in the following steps S37a to S37d.

[0108] S37a. Convert the contrast-enhanced image into a Lab image and calculate the channel mean of the three channels in the Lab image; after calculating the channel mean of the three channels in the Lab image, the contrast weight can be calculated, as shown in step S37b below.

[0109] S37b. Based on the channel mean of the three channels and the Lab value of each pixel in the Lab image, calculate the contrast weight of each pixel in the contrast-enhanced image. In specific implementation, for any pixel in the contrast-enhanced image, first calculate the first difference between the L channel value and the L channel mean of that pixel in the Lab image, the second difference between the a channel value and the a channel mean, and the third difference between the b channel value and the b channel mean. Then, calculate the square of the first difference, the square of the second difference, and the square of the third difference. Finally, sum the three squared results to obtain the contrast weight of that pixel. In this way, the weight, combined with color and brightness features, is used to calculate the salient areas in the image, which can highlight targets with low recognition rates in underwater scenes.

[0110] After the contrast weight is calculated, the color weight and exposure weight can be calculated, as shown in step S37c below.

[0111] S37c. Based on the RGB values ​​of each pixel in the contrast-enhanced image, calculate the color weight and exposure weight of each pixel in the contrast-enhanced image; in specific implementation, for any pixel in the contrast-enhanced image, for example, but not limited to, the following formula can be used to calculate the color weight of that pixel.

[0112] ; In the formula, This represents the color weight of any given pixel. These are the R channel value, G channel value, and B channel value of any given pixel. This represents the brightness value of any given pixel, which is calculated from the RGB values ​​of that pixel. .

[0113] Similarly, the formula for calculating the exposure weight of any pixel is: ; In the formula, This represents the exposure weight of any pixel.

[0114] Thus, based on the aforementioned formula, the color weight and exposure weight of each pixel in the contrast-enhanced image can be calculated; then, the first image fusion weight of each pixel can be obtained by combining the contrast weight, as shown in step S37d below.

[0115] S37d. Using the contrast weight, color weight, and exposure weight of each pixel in the contrast-enhanced image, the first image fusion weight of each pixel is determined; in this embodiment, the first image fusion weight of each pixel is obtained by summing the contrast weight, color weight, and exposure weight of each pixel; thus, the first image fusion weight of all pixels can be used to form a first initial feature weight map.

[0116] After generating the first initial feature weight map through the aforementioned steps S37a to S37d, the second initial feature weight map corresponding to the detail enhancement image can be generated in the same way, as shown in step S38 below.

[0117] S38. Determine the second image fusion weight of each pixel in the detail-enhanced image, and generate a second initial feature weight map corresponding to the detail-enhanced image based on the second image fusion weight of each pixel in the detail-enhanced image; in this embodiment, the generation process of the second initial feature weight map can be referred to step S37, and will not be repeated here.

[0118] After obtaining the two initial feature weight maps, regularization can be performed, as shown in step S39 below.

[0119] S39. Regularize the first initial feature weight map and the second initial feature weight map to obtain the first feature weight map and the second feature weight map; in this embodiment, the following formula is used to calculate the first feature weight map.

[0120] ; In the formula, This represents the first feature weight map. This represents the first initial feature weight map. Represents the regularization coefficient. This represents the initial feature weight map of the u-th feature.

[0121] After calculating the first feature weight map and the second feature weight map as described above, image fusion can be performed, as shown in step S310 below.

[0122] S310. Using the first feature weight map and the second feature weight map, perform image fusion on the contrast-enhanced image and the detail-enhanced image to obtain the enhanced leaf image after image fusion; in specific applications, this embodiment provides a multi-scale fusion strategy, the process of which is shown in the following steps S310a to S310d.

[0123] S310a. Construct a first multi-level pyramid for each of the contrast-enhanced image and the detail-enhanced image, wherein the first multi-level pyramid includes multiple layers of images, and each layer of images has a different resolution; in a specific implementation, the construction process of the first multi-level pyramid is exemplified as follows: Gaussian blur and downsampling operations are performed on the contrast-enhanced image to construct a Gaussian pyramid; then, starting from the top layer of the Gaussian pyramid, the first multi-level pyramid is constructed by upsampling the image of the next layer and subtracting it from the image of the current layer until the bottom layer image of the Gaussian pyramid is polled.

[0124] After obtaining the first multi-level pyramids of the contrast-enhanced image and the detail-enhanced image respectively, the multi-level pyramids of the first feature weight map and the second feature weight map can be constructed, as shown in step S310b below.

[0125] S310b. Construct the second multi-level pyramid corresponding to the first feature weight map and the second feature weight map respectively; in this embodiment, the aforementioned second multi-level pyramid is exemplified by the Gaussian pyramid, and its construction process has been described above and will not be repeated here; thus, after the construction of the multi-level pyramid is completed, image fusion can be performed, and the process is as shown in the following step S310c.

[0126] S310c. The images of the two first multi-level pyramids and the two second multi-level pyramids at each level are fused to obtain a multi-level fused image pyramid. In this embodiment, the contrast-enhanced image at level t is multiplied by the image of the first feature weight map at level t (i.e., the weight value), and then the detail-enhanced image at level t is multiplied by the image of the second feature weight map at level t (i.e., the weight value) to obtain two sub-images. Finally, the two sub-images are added together to obtain the fused image at level t. Thus, based on the aforementioned method, a multi-level fused image pyramid can be obtained. Finally, the fused image pyramid can be upsampled to reconstruct the enhanced leaf image, as shown in step S310d below.

[0127] S310d. Starting from the top layer of the fused image pyramid, image upsampling is performed until the bottom layer of the fused image pyramid is upsampled to obtain the enhanced leaf image.

[0128] Thus, through the aforementioned steps S310a to S310d, this embodiment adopts a multi-scale fusion strategy to fuse features from different input images and ensure the effectiveness of each feature weight map. As a result, compared to simple linear weighted fusion, the image fusion effect is better and more conducive to subsequent coating wear identification.

[0129] After obtaining the enhanced leaf image, image segmentation can be performed, as shown in step S4 below.

[0130] S4. Perform image segmentation processing on the enhanced blade image to obtain the coating wear area; in this embodiment, first perform edge detection processing on the enhanced blade image to obtain an edge detection image; then, perform morphological operations on the edge detection image to obtain the coating wear area after the morphological operations; of course, edge detection and morphological operations are common techniques for image segmentation, and will not be elaborated here.

[0131] After obtaining the wear area of ​​the coating, feature extraction can be performed, as shown in step S5 below.

[0132] S5. Perform feature extraction processing on the coating wear area to obtain the texture features and geometric features of the blade image, and construct the coating wear features using the texture features and geometric features. In this embodiment, for example, but not limited to, first calculate the gray-level co-occurrence matrix of the coating wear area; then, determine several texture statistics (such as contrast, correlation, homogeneity, etc.) based on the gray-level co-occurrence matrix; next, calculate the LBP histogram of the coating wear area (which is not sensitive to changes in illumination and has a strong texture description ability); finally, use the several texture statistics and the LBP histogram to generate the texture features.

[0133] Similarly, for geometric features, the centroid coordinates, area, relative area, circularity, rectangularity, and triangularity of the coating wear area can be extracted; then, the centroid coordinates, area, relative area, circularity, rectangularity, and triangularity are used to form the geometric features; of course, the aforementioned centroid coordinates, area, relative area, circularity, rectangularity, and triangularity are commonly used geometric features, and their calculation principles will not be elaborated here.

[0134] Thus, after extracting geometric and texture features, they can be normalized and then spliced ​​together to form coating wear features. Finally, the coating wear recognition result can be obtained by using a pre-trained coating wear recognition model, as shown in step S6 below.

[0135] S6. Input the coating wear characteristics into the coating wear identification model to obtain the coating wear identification result of the turbine blades. In this embodiment, the coating wear identification model can be, but is not limited to, a trained SVM model, and is trained by taking the sample coating wear characteristics of several sample blades as input and the coating wear identification result of each sample blade as output (the result is the wear type, such as peeling, cracks, etc.). Finally, after training, the coating wear identification model can be obtained. In actual use, the coating wear characteristics obtained in step S5 above can be directly input into the model to obtain the coating wear identification result of the turbine blades. Based on this, an early warning can be sent to the turbine maintenance end according to the coating wear identification result of the turbine blades to remind the staff to perform timely maintenance of the turbine blades.

[0136] Therefore, the turbine composite coating wear visualization and identification method described in detail in steps S1 to S6 above, compared with traditional manual inspection, not only improves efficiency but also improves the accuracy of detection because wear identification does not rely on the professional experience of personnel. Furthermore, during inspection, the actual working environment of the turbine blades is considered, and targeted processing such as deblurring, color correction, contrast enhancement, and detail enhancement is applied to the blade images. This reduces the interference of the complex underwater environment on blade coating wear detection, thus solving the limitations of automated coating wear identification applications. It also reduces noise interference in coating wear identification, further improving the accuracy of identification. Therefore, this invention provides a highly efficient and accurate automatic coating wear detection technology, filling the technological gap in the field of automatic wear detection of turbine blade coatings, and is therefore very suitable for large-scale application and promotion.

[0137] like Figure 2As shown, the third aspect of this embodiment provides a hardware system for implementing the visual identification method for wear of composite coatings in water turbines as described in the first aspect of the embodiment, comprising: The calibration unit is used to acquire images of the turbine blades and determine the calibration color information of the blade images.

[0138] The image correction unit is used to perform deblurring and color correction processing on the leaf image according to the calibration color information to obtain the corrected image.

[0139] The image enhancement unit is used to perform contrast enhancement processing and detail enhancement processing on the corrected image respectively to obtain a contrast-enhanced image and a detail-enhanced image, and to perform image fusion on the contrast-enhanced image and the detail-enhanced image to obtain an enhanced leaf image.

[0140] An image segmentation unit is used to perform image segmentation processing on the enhanced blade image to obtain the coating wear area.

[0141] The feature extraction unit is used to perform feature extraction processing on the coating wear area to obtain the texture features and geometric features of the blade image, and to construct the coating wear features using the texture features and geometric features.

[0142] The wear identification unit is used to input the coating wear characteristics into the coating wear identification model to obtain the coating wear identification result of the turbine blade.

[0143] The working process, working details and technical effects of the system provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0144] like Figure 3 As shown, the third aspect of this embodiment provides a device for visually identifying wear of composite coatings on water turbines. Taking the device as an electronic device as an example, it includes: a memory, a processor, and a transceiver connected in sequence. The memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the method for visually identifying wear of composite coatings on water turbines as described in the first aspect of the embodiment.

[0145] For specific examples, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0146] In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. For example, the processor may not be limited to microprocessors of the STM32F105 series, reduced instruction set computer (RISC) microprocessors, x86 architecture processors, or processors with integrated neural network processing units (NPUs). The transceiver may be, but is not limited to, a Wi-Fi transceiver, a Bluetooth transceiver, a General Packet Radio Service (GPRS) transceiver, a ZigBee (a low-power LAN protocol based on the IEEE 802.15.4 standard) transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver. Furthermore, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0147] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0148] The fourth aspect of this embodiment provides a storage medium that stores instructions containing the method for visually identifying wear of composite coatings on water turbines as described in the first aspect of this embodiment. That is, the storage medium stores instructions that, when executed on a computer, perform the method for visually identifying wear of composite coatings on water turbines as described in the first aspect of this embodiment.

[0149] The storage medium refers to a carrier for storing data, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or memory sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0150] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0151] The fifth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the method for visually identifying wear of composite coatings on water turbines as described in the first aspect of this embodiment. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0152] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for visually identifying wear on composite coatings of hydraulic turbines, characterized in that, include: Acquire images of the turbine blades and determine the calibration color information of the blade images; Based on the calibrated color information, the leaf image is deblurred and color corrected to obtain the corrected image; The corrected image is subjected to contrast enhancement processing and detail enhancement processing respectively to obtain a contrast-enhanced image and a detail-enhanced image. The contrast-enhanced image and the detail-enhanced image are then fused to obtain an enhanced leaf image. The enhanced blade image is segmented to obtain the coating wear area; Feature extraction processing is performed on the wear area of ​​the coating to obtain the texture features and geometric features of the blade image, and the coating wear features are constructed using the texture features and geometric features; The coating wear characteristics are input into the coating wear identification model to obtain the coating wear identification results of the turbine blades.

2. The method for visually identifying wear of composite coatings in water turbines according to claim 1, characterized in that, Determining the calibration color information of the blade image includes: Initialize the number of searches i, and obtain the target image at the time of the i-th search, where the initial value of i is 1, and when i is 1, the target image at the time of the i-th search is the leaf image; The target image is divided into equal parts to obtain several image blocks; Calculate the pixel variance of each image patch and select the image patch with the largest pixel variance as the optimal image patch for the i-th search. Determine whether the search stopping condition is met, wherein the search stopping condition is that the total number of pixels in the optimal image block is less than or equal to a pixel count threshold; If not, the optimal image patch in the i-th search will be used as the target image in the (i+1)-th search. Increment i by 1 and reacquire the target image at the time of the i-th search until the search stopping condition is met, so that the optimal image block at the time of the search stopping condition is used as the calibration image block; The calibration color information is determined based on the calibration image block.

3. The method for visually identifying wear of composite coatings in hydraulic turbines according to claim 1, characterized in that, Based on the calibrated color information, the leaf image is deblurred and color corrected to obtain a corrected image, including: Based on the calibrated color information, the blur correction factor of the leaf image in different channels is determined; The leaf image is deblurred using the blur correction factors on different channels and the calibrated color information to obtain a deblurred image; Based on the calibrated color information, the color compensation type of the deblurred image is determined; Based on the color compensation type of the deblurred image, determine the color compensation parameters; Based on the color compensation parameters, the deblurred image is subjected to color correction processing to obtain the corrected image.

4. The method for visually identifying wear of composite coatings in water turbines according to claim 3, characterized in that, The calibration color information includes the calibration colors of the leaf image on three color channels, and the three color channels include the R channel, G channel and B channel; Specifically, based on the calibrated color information, the blur correction factors for the leaf image in different channels are determined, including: Based on the calibrated colors in the three color channels, the channel values ​​of each pixel in the leaf image are normalized in the three color channels to obtain the normalized channel values ​​of each pixel in the three color channels. For any color channel, the smallest normalized channel value is selected from the normalized channel values ​​of each pixel in that color channel and used as the calibration channel value. After all color channels have been queried, three calibration channel values ​​are obtained. The smallest calibration channel value is selected from the three calibration channel values, and the fuzzy correction factor corresponding to the R channel is calculated based on the smallest calibration channel value. The first and second deblurring factors are calculated based on the calibrated colors in the three color channels. The fuzz correction factor for the G channel is determined using the fuzz correction factor and the first defuzzification factor corresponding to the R channel, and the fuzz correction factor for the B channel is determined using the fuzz correction factor and the second defuzzification factor corresponding to the R channel.

5. The method for visually identifying wear of composite coatings in water turbines according to claim 1, characterized in that, The corrected image undergoes contrast enhancement processing, including: The leaf image and the corrected image are converted to HSV space respectively to obtain the first HSV image and the second HSV image; Calculate the first mean and first standard deviation of the S channel in the first HSV image, and the second mean and second standard deviation of the V channel in the first HSV image; Calculate the third mean and third standard deviation of the S channel in the second HSV image, and the fourth mean and fourth standard deviation of the V channel in the second HSV image; The contrast enhancement factor of the S channel is calculated based on the first mean, the third mean, the first standard deviation, and the third standard deviation; and the contrast enhancement factor of the V channel is calculated based on the second mean, the fourth mean, the second standard deviation, and the fourth standard deviation. By using the contrast enhancement factors of the S channel and the V channel, the contrast of the S channel and V channel in the second HSV image is enhanced to obtain an enhanced HSV image. The enhanced HSV image is converted to RGB space to obtain the contrast-enhanced image.

6. The method for visually identifying wear of composite coatings in hydraulic turbines according to claim 1, characterized in that, Image fusion is performed on the contrast-enhanced image and the detail-enhanced image to obtain an enhanced leaf image, including: The first image fusion weight of each pixel in the contrast-enhanced image is determined, and a first initial feature weight map corresponding to the contrast-enhanced image is generated based on the first image fusion weight of each pixel in the contrast-enhanced image. The second image fusion weight of each pixel in the detail-enhanced image is determined, and a second initial feature weight map corresponding to the detail-enhanced image is generated based on the second image fusion weight of each pixel in the detail-enhanced image. The first initial feature weight map and the second initial feature weight map are regularized to obtain the first feature weight map and the second feature weight map. Using the first feature weight map and the second feature weight map, the contrast-enhanced image and the detail-enhanced image are fused to obtain the enhanced leaf image after image fusion.

7. The method for visually identifying wear of composite coatings in a water turbine according to claim 6, characterized in that, The first image fusion weight for each pixel in the contrast-enhanced image is determined, including: The contrast-enhanced image is converted into a Lab image, and the channel mean of the three channels in the Lab image is calculated. Based on the channel mean of the three channels and the Lab value of each pixel in the Lab image, the contrast weight of each pixel in the contrast-enhanced image is calculated. Based on the RGB values ​​of each pixel in the contrast-enhanced image, the color weight and exposure weight of each pixel in the contrast-enhanced image are calculated. The first image fusion weight of each pixel is determined by using the contrast weight, color weight, and exposure weight of each pixel in the contrast-enhanced image.

8. The method for visually identifying wear of composite coatings in a water turbine according to claim 6, characterized in that, Using the first feature weight map and the second feature weight map, image fusion is performed on the contrast-enhanced image and the detail-enhanced image, including: A first multi-level pyramid is constructed for the contrast-enhanced image and the detail-enhanced image, wherein the first multi-level pyramid includes multiple layers of images, and each layer of images has a different resolution; Construct the second multi-level pyramid corresponding to the first feature weight map and the second feature weight map respectively; By fusing the images of the two first-level multi-level pyramids and the two second-level multi-level pyramids at each level, a multi-level fused image pyramid is obtained. The enhanced leaf image is obtained by upsampling the image from the top layer of the fused image pyramid until the bottom layer of the fused image pyramid has been upsampled.

9. The method for visually identifying wear of composite coatings in water turbines according to claim 1, characterized in that, The enhanced blade image is segmented to obtain the coating wear area, including: Edge detection processing is performed on the enhanced blade image to obtain an edge-detected image; Morphological operations are performed on the edge detection image to obtain the coating wear area after the morphological operations; Correspondingly, feature extraction processing for the worn areas of the coating includes: Calculate the gray-level co-occurrence matrix of the wear area of ​​the coating; Based on the gray-level co-occurrence matrix, several texture statistics are determined; Calculate the LBP histogram of the wear area of ​​the coating; The texture features are generated using several texture statistics and the LBP histogram; Extract the centroid coordinates, area, relative area, circularity, rectangularity, and triangularity of the wear area of ​​the coating; The geometric features are constructed using centroid coordinates, area, relative area, circularity, rectangularity, and triangularity.

10. A visual identification system for wear of composite coatings on water turbines, characterized in that, include: The calibration unit is used to acquire images of the turbine blades and determine the calibration color information of the blade images. The image correction unit is used to perform deblurring and color correction processing on the leaf image according to the calibration color information to obtain the corrected image; An image enhancement unit is used to perform contrast enhancement processing and detail enhancement processing on the corrected image respectively to obtain a contrast-enhanced image and a detail-enhanced image, and to perform image fusion on the contrast-enhanced image and the detail-enhanced image to obtain an enhanced leaf image; An image segmentation unit is used to perform image segmentation processing on the enhanced blade image to obtain the coating wear area; The feature extraction unit is used to perform feature extraction processing on the coating wear area to obtain the texture features and geometric features of the blade image, and to construct the coating wear features using the texture features and the geometric features. The wear identification unit is used to input the coating wear characteristics into the coating wear identification model to obtain the coating wear identification result of the turbine blade.

Images