A kind of image recognition method and system of picking net net wear

Abstract

The present application relates to the technical field of image recognition, in particular to a kind of image recognition method and system for wear of twining net.The present application, by collecting the original underwater image of twining net in underwater operation time period, and determining net clothes structure layer and water background layer from original underwater image, make net clothes own structure information and water brightness distribution information separate processing;Then determine scattering light spot binary mask, and determine biological attachment layer by color space conversion and texture difference judgment, so that local highlight interference and biological attachment area can be individually referred in subsequent processing;Then refer to scattering light spot binary mask and water background layer to generate optimized underwater image, and then refer to biological attachment layer, net clothes structure layer and optimized underwater image to reconstruct net clothes clear structure form;Finally, determine net line continuous line pixel set from reconstructed net clothes image, which plays a role of connecting underwater interference processing process and net clothes wear determination process.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method and system for recognizing the wear of mesh fabric in a sifting net. Background Technology

[0002] The image recognition method for the wear of fishing nets refers to the process of image analysis and judgment of the wear state of the net surface of fishing nets. This method is mainly used to monitor the physical damage of the net during actual use in order to assess the usability of the relevant fishing gear. Typically, a camera device is used to acquire a visual image of the net surface and traditional image processing methods are used to extract the texture edge features of the net for comparison and judgment.

[0003] Traditional methods for identifying wear on scavenging nets, while capable of acquiring visual images of the net surface using cameras and determining wear status through extraction and comparison of net texture edge features, face several challenges. When the scavenging net is operating underwater, water scattering in the original underwater image alters the brightness distribution of the net area, fogging weakens the clarity of the net line boundaries, scattered light spots create abnormally bright areas, and biofouling covers or alters the original texture of the net. Therefore, when directly processing underwater images, traditional methods easily misinterpret changes in the water background as changes in the net texture, and also easily mistake irregular areas formed by biofouling as worn areas. This leads to difficulties in accurately distinguishing between net line continuity, gaps between net lines, variations in net line width, and texture deviations. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose a method and system for image recognition of wear on the mesh of a snagging net.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for identifying the wear and tear of underwater sorting nets, used to identify the wear and tear of the nets during underwater sorting net operations, comprising the following steps: Collect raw underwater images of the sifting net during its underwater operation period and determine the net structure layer and water background layer in the raw underwater images; The binary mask of scattered light spots in the original underwater image is determined, and the original underwater image is converted from RGB color space to HSV color space to determine the bio-attachment layer in the original underwater image. By referring to the binary mask of scattered light spots and the water background layer, the interference of water scattering and fogging in the original underwater image is eliminated, and the optimized underwater image is generated. By referring to the bio-attached layer, the net structure layer and the optimized underwater image, the clear structural morphology of the net without bio-attachment and water disturbance is reconstructed, and the reconstructed net image is generated. Determine the set of pixels of continuous lines of mesh in the reconstructed mesh image, and calculate the gap length of mesh fibers based on the set of pixels of continuous lines of mesh; Scan the pixel set of continuous lines of the mesh, calculate the percentage reduction in the width of the mesh fibers and the texture feature deviation of the mesh fibers, and combine the gap length of the mesh fibers to determine whether there are gaps, width reduction and texture feature deviation of the mesh fibers, and generate the image recognition result of mesh wear of the snagging mesh.

[0006] A system for recognizing the wear of underwater sorting nets is provided for identifying the wear condition of the nets during underwater sorting net operations. The system includes: The image layering module acquires raw underwater images of the sifting net during its underwater operation period and determines the net structure layer and water background layer in the raw underwater images. The interference feature extraction module determines the binary mask of scattered light spots in the original underwater image, and converts the original underwater image from RGB color space to HSV color space to determine the biological attachment layer in the original underwater image. The background optimization module, referencing the binary mask of scattered light spots and the water background layer, eliminates water scattering and fogging blur interference in the original underwater image and generates an optimized underwater image. The image reconstruction module, referring to the bio-attached layer, the net structure layer and the optimized underwater image, reconstructs a clear structural morphology of the net without bio-attachment and water disturbance, and generates a reconstructed net image. The line feature extraction module determines the set of pixels of continuous lines in the reconstructed mesh image and calculates the gap length of the mesh fibers based on the set of pixels of continuous lines. The mesh wear recognition module scans the pixel set of continuous lines of the mesh, calculates the percentage reduction in the width of the mesh fibers and the deviation in the texture features of the mesh fibers, and combines the gap length of the mesh fibers to determine whether there are gaps, width reduction and texture feature deviation of the mesh fibers, and generates the mesh wear image recognition result of the sifting net.

[0007] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, original underwater images of the scavenging net during its underwater operation are acquired, and the net structure layer and water background layer are determined from these images, allowing the net's structural information to be processed separately from the water brightness distribution information. Then, a binary mask of scattered light spots is determined, and the bio-attachment layer is identified through color space conversion and texture difference judgment, enabling localized high-brightness interference and bio-attachment areas to be referenced separately in subsequent processing. Next, an optimized underwater image is generated by referring to the binary mask of scattered light spots and the water background layer, and then the clear structural morphology of the net is reconstructed by referring to the bio-attachment layer, the net structure layer, and the optimized underwater image. Finally, the set of continuous line pixels of the net is determined from the reconstructed net image, and the gap length, width reduction percentage, and texture feature deviation of the net fibers are further calculated. This allows for the generation of scavenging net wear image recognition results based on the presence of gaps, width reduction, and texture feature deviation in the net fibers, effectively connecting the underwater interference processing process with the net wear determination process. Attached Figure Description

[0008] Figure 1 This is a flowchart of the method steps of the present invention; Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0010] Please see Figure 1 This invention provides a technical solution: a method for identifying the wear and tear of underwater sorting nets, used to identify the wear and tear of the nets during underwater sorting net operations, comprising the following steps: S1: Acquire raw underwater images of the winch net during its underwater operation period, and determine the net structure layer and water background layer in the raw underwater images, including: Collect raw underwater images of the sorting net during the operation period, calculate the pixel intensity gradient value of each raw pixel in the raw underwater image with its horizontal and vertical adjacent raw pixels, and calculate the gradient magnitude of each raw pixel based on the horizontal and vertical pixel intensity gradient values. After acquiring raw underwater images of the winch net during its operation period, the pixel intensity of each raw pixel is read point-by-point according to the spatial coordinates of the raw underwater image. The raw underwater image is the image formed by the winch net during its underwater operation period, and the raw pixel is the image location in the raw underwater image with spatial coordinates and pixel intensity. The pixel intensity is the brightness value corresponding to the raw pixel. When the raw pixel is located at a non-boundary position in the horizontal direction, the pixel intensity of the raw pixels adjacent to it in the horizontal direction is read; when the raw pixel is located at a boundary position in the horizontal direction, the pixel intensity of the raw pixels inside the horizontal boundary is read. When the raw pixel is located at a non-boundary position in the vertical direction, the pixel intensity of the raw pixels adjacent to it in the vertical direction is read; when the raw pixel is located at a boundary position in the vertical direction, the pixel intensity of the raw pixels inside the vertical boundary is read. The pixel intensity gradient value in the horizontal direction is equal to the pixel intensity difference between the raw pixels adjacent to it in the horizontal direction, and the pixel intensity gradient value in the vertical direction is equal to the pixel intensity difference between the raw pixels adjacent to it in the vertical direction. Gradient magnitude = intensity change formed by the pixel intensity gradient values ​​in the horizontal and vertical directions. When pixel intensity gradient values ​​exist in both the horizontal and vertical directions, the pixel intensity gradient values ​​in both directions are used as the basis for calculating the current gradient magnitude. When only pixel intensity gradient values ​​exist in the horizontal direction, the pixel intensity gradient value in the horizontal direction is used as the basis for calculating the current gradient magnitude. When only pixel intensity gradient values ​​exist in the vertical direction, the pixel intensity gradient value in the vertical direction is used as the basis for calculating the current gradient magnitude, thus obtaining the gradient magnitude of each original pixel.

[0011] Each original pixel in the original underwater image is regarded as the central original pixel. The pixel neighborhood window is divided, and the variance of the mesh pixel intensity within each pixel neighborhood window is calculated. The gradient magnitude of each original pixel and the variance of the mesh pixel intensity are normalized respectively. The weighted sum of the normalized gradient magnitude and the normalized mesh pixel intensity variance is calculated to obtain the mesh structure feature value. Each original pixel in the raw underwater image is considered the central original pixel, and a pixel neighborhood window is defined around the central original pixel. The pixel neighborhood window represents the local range around the central original pixel from which the intensity of surrounding pixels is read. The size of the pixel neighborhood window is set based on the local pixel span of the net structure in the raw underwater image. The local pixel span of the net structure is obtained by statistically analyzing the number of original pixels spanned by adjacent brightness changes in the net structure in the raw underwater image. When the pixel neighborhood window extends beyond the boundary of the raw underwater image, the nearest original pixel within the boundary is used as the source of values ​​for the missing positions in the pixel neighborhood window. The variance of the net pixel intensity within each pixel neighborhood window is obtained from the pixel intensity of all original pixels within the pixel neighborhood window. The variance of the net pixel intensity represents the degree of dispersion of the pixel intensity around the average value within the pixel neighborhood window. The average pixel intensity within the pixel neighborhood window = the sum of the pixel intensities of all original pixels within the pixel neighborhood window / the number of original pixels within the pixel neighborhood window; the variance of the net pixel intensity = the sum of the squares of the deviations between the pixel intensity of each original pixel within the pixel neighborhood window and the average pixel intensity / the number of original pixels within the pixel neighborhood window. When the maximum gradient magnitude is greater than the minimum gradient magnitude, the normalized gradient magnitude = (current original pixel gradient magnitude - minimum gradient magnitude) / (maximum gradient magnitude - minimum gradient magnitude); when the maximum gradient magnitude equals the minimum gradient magnitude, the normalized gradient magnitude is set to 0. When the maximum netting pixel intensity variance is greater than the minimum netting pixel intensity variance, the normalized netting pixel intensity variance = (current netting pixel intensity variance - minimum netting pixel intensity variance) / (maximum netting pixel intensity variance - minimum netting pixel intensity variance); when the maximum netting pixel intensity variance equals the minimum netting pixel intensity variance, the normalized netting pixel intensity variance is set to 0. The weights in the weighted sum are set according to the proportion of the changes in the normalized gradient magnitude and the normalized netting pixel intensity variance at prominent locations within the netting structure. The changes in the normalized gradient magnitude and the normalized netting pixel intensity variance are obtained statistically from the distribution of values ​​corresponding to prominent locations within the netting structure. The feature value of the mesh structure is obtained by calculating the normalized gradient magnitude × the weight corresponding to the gradient magnitude + the normalized variance of the mesh pixel intensity × the weight corresponding to the variance of the mesh pixel intensity.

[0012] The feature values ​​of the mesh structure corresponding to each original pixel are arranged according to the spatial coordinate distribution of the original underwater image to generate a mesh structure layer. At the same time, the pixel intensity of the central original pixel and the corresponding surrounding original pixels in the original underwater image are weighted and averaged. Based on the calculation results, the edge details in the original underwater image are smoothed and the brightness distribution information of the water body is preserved to generate a water background layer. The feature value of the mesh structure corresponding to each original pixel is read and arranged according to the spatial coordinate distribution of the original underwater image. The spatial coordinate distribution is the arrangement relationship of the horizontal and vertical coordinates of the original pixels in the original underwater image. The mesh structure layer is a data layer that stores the mesh structure feature values ​​according to the spatial coordinates of the original underwater image. The mesh structure feature value of the original pixel is written to the spatial coordinate position corresponding to the original pixel. When calculating the weighted average of the pixel intensity of the central original pixel and its corresponding surrounding original pixels in the original underwater image, the pixel intensity of the central original pixel and the pixel intensity of its corresponding surrounding original pixels are read. The corresponding surrounding original pixels are the original pixels that participate in local smoothing around the central original pixel. The weights in the weighted average calculation are set according to the spatial distance of the central original pixel and its corresponding surrounding original pixels relative to the central original pixel. The spatial distance corresponding to the central original pixel is recorded as 0, and the spatial distance corresponding to the surrounding original pixels is determined by the difference in horizontal and vertical coordinates. The weight corresponding to the original pixel = 1 / (the spatial distance corresponding to the original pixel + 1), and the weight corresponding to the central original pixel = 1. The weighted average calculation result is calculated as follows: (Pixel intensity of the central original pixel × Weight of the central original pixel + Sum of pixel intensity of each corresponding surrounding original pixel × Weight of each corresponding surrounding original pixel) / (Sum of weight of the central original pixel and weight of each corresponding surrounding original pixel). When all surrounding original pixels are complete, the central original pixel and all corresponding surrounding original pixels participate in the weighted average calculation. When surrounding original pixels are missing due to boundary issues, the central original pixel and all valid surrounding original pixels within the boundary participate in the weighted average calculation. The weighted average calculation result is written to the coordinates of the central original pixel to generate the water background layer.

[0013] S2: Determine the binary mask of the scattered light spot in the original underwater image, and simultaneously convert the original underwater image from RGB color space to HSV color space. Determine the biofilm layer in the original underwater image, including: For each central original pixel in the original underwater image, an analysis neighborhood is divided. The peak pixel intensity in each analysis neighborhood is detected. The difference between the peak pixel intensity and the preset global brightness threshold is calculated to filter out pixels with abnormal brightness in the water body. At the same time, a binary mask matrix with the same size as the original underwater image is constructed. The mask value of the coordinate position of the pixel with abnormal brightness in the binary mask matrix is ​​marked as the value one, and the mask value of the coordinate position of the original pixel that is not identified as a pixel with abnormal brightness is marked as the value zero. The marked binary mask matrix is ​​used as a binary mask for scattered light spots. For each central original pixel in the original underwater image, an analysis neighborhood is defined. The analysis neighborhood is the local area surrounding the central original pixel where pixel intensity peaks are detected. The size of the analysis neighborhood is pre-defined based on the continuous spatial distribution range of pixels with abnormal water brightness in the original underwater image. This continuous spatial distribution range is obtained by counting the number of neighboring locally bright pixels in the original underwater image. When the analysis neighborhood extends beyond the boundary of the original underwater image, valid original pixels within the boundary participate in pixel intensity peak detection. Pixel intensity peak = the maximum pixel intensity of all original pixels within the analysis neighborhood. The preset global brightness threshold is set based on the pixel intensity statistics of the water background pixels in the water background layer. First, the pixel intensities of all water background pixels in the water background layer are arranged from low to high. Then, the average pixel intensity and standard deviation of all water background pixels in the water background layer are calculated. The preset global brightness threshold = average pixel intensity of all water background pixels in the water background layer + 3 × standard deviation of pixel intensity of all water background pixels in the water background layer. When the preset global brightness threshold is greater than the maximum pixel intensity of all water background pixels in the water background layer, the maximum pixel intensity of all water background pixels in the water background layer is used as the preset global brightness threshold. The difference = peak pixel intensity - preset global brightness threshold. When the difference is greater than 0, the corresponding coordinate position is determined to be a pixel with abnormal brightness in the water; when the difference is equal to 0 or less than 0, the corresponding coordinate position is not determined to be a pixel with abnormal brightness in the water. The binarized mask matrix is ​​a matrix with the same size as the original underwater image and stores values ​​of 1 and 0 according to the coordinates of the original underwater image. The mask value of the coordinate position of the pixel with abnormal brightness in the water body is marked as 1, and the mask value of the coordinate position of the original pixel that is not identified as a pixel with abnormal brightness is marked as 0. The marked binarized mask matrix is ​​used as the binary mask of the scattered light spot.

[0014] The original underwater image is converted from RGB color space to HSV color space. Hue and saturation in HSV color space are extracted. When the hue is within the preset range of marine life hue and the saturation is greater than the preset saturation threshold, the original pixels corresponding to the hue and saturation are retained. When the hue exceeds the preset range of marine life hue or the saturation is less than or equal to the preset saturation threshold, the original pixels corresponding to the hue and saturation are removed to obtain the color segmentation region. When converting the original underwater image from RGB color space to HSV color space, the color channel values ​​of each original pixel in the original underwater image in RGB color space are read and converted into hue and saturation in HSV color space. RGB color space is the color space that records the color channel values ​​in the original underwater image, while HSV color space is a color space that expresses color through hue, saturation, and lightness. Hue is the color category value, and saturation is the color purity value. The preset marine organism hue range is set according to the hue of the original pixels corresponding to the organism attachment morphology in the original underwater image. First, the original pixels corresponding to the organism attachment morphology are selected in the original underwater image. The hue of the selected original pixels in the HSV color space is read. The hues of the selected original pixels are arranged from low to high values. The 5th percentile hue after arrangement is taken as the lower limit of the preset marine organism hue range, and the 95th percentile hue after arrangement is taken as the upper limit of the preset marine organism hue range. When the lower limit of the preset marine organism hue range is less than the upper limit, the closed range between the two is taken as the preset marine organism hue range. When the lower limit of the preset marine organism hue range is equal to the upper limit, the same hue is taken as the single-value hue range. The preset saturation threshold is set according to the saturation of the original pixels corresponding to the biological attachment morphology and the original pixels corresponding to the water background. First, the minimum saturation value of the original pixels corresponding to the biological attachment morphology is read, and then the maximum saturation value of the original pixels corresponding to the water background is read. When the minimum saturation value of the original pixels corresponding to the biological attachment morphology is greater than the maximum saturation value of the original pixels corresponding to the water background, the preset saturation threshold is calculated as (minimum saturation value of the original pixels corresponding to the biological attachment morphology + maximum saturation value of the original pixels corresponding to the water background) / 2. When the minimum saturation value of the original pixels corresponding to the biological attachment morphology is less than or equal to the maximum saturation value of the original pixels corresponding to the water background, the saturation values ​​of the original pixels corresponding to the biological attachment morphology and the original pixels corresponding to the water background are merged and sorted. The median value of two adjacent saturation values ​​is taken as the candidate threshold. The number of original pixels corresponding to the water background that are retained and the number of original pixels corresponding to the biological attachment morphology that are removed are counted under the candidate threshold. The candidate threshold with the smallest sum of the two is taken as the preset saturation threshold. When the hue is within the preset range of marine life hues and the saturation is greater than the preset saturation threshold, the original pixels corresponding to the hue and saturation are retained; when the hue is within the preset range of marine life hues and the saturation is less than or equal to the preset saturation threshold, the original pixels corresponding to the hue and saturation are discarded; when the hue exceeds the preset range of marine life hues, the original pixels corresponding to the hue and saturation are discarded to obtain the color segmentation region.

[0015] The original underwater image is converted into a grayscale image. In the grayscale image, the grayscale value of each original pixel is compared with that of its corresponding neighboring original pixels. If the grayscale value of the neighboring original pixel is greater than or equal to that of the original pixel, it is recorded as one; otherwise, it is recorded as zero. The values ​​corresponding to each neighboring original pixel are combined into a binary sequence and converted into decimal local binary feature values. The local binary feature values ​​of each original pixel in the color segmentation region are counted to generate the corresponding mesh texture histogram. The difference between the mesh texture histogram and the preset uniform weave structure reference histogram is calculated. When the difference is greater than the preset morphological difference threshold, the corresponding original pixel region is determined to be an irregular morphological region. All irregular morphological regions are extracted and mapped according to the spatial coordinates of the original underwater image to generate a bio-attachment layer. When converting the original underwater image to a grayscale image, the color channel values ​​of each original pixel in the original underwater image in the RGB color space are read and converted into grayscale values. The grayscale image is an image that stores grayscale values ​​according to the spatial coordinates of the original underwater image. Centered on each original pixel, the grayscale values ​​of corresponding adjacent original pixels are read. The adjacent original pixels are the neighboring pixel positions surrounding the original pixel. A grayscale value greater than or equal to the grayscale value of the original pixel is recorded as one, and a grayscale value less than the original pixel is recorded as zero. The values ​​corresponding to each adjacent original pixel are combined into a binary sequence according to their adjacent positions and converted into decimal local binary feature values. The local binary feature value is a numerical expression of the arrangement relationship of grayscale values ​​around the original pixel. The mesh texture histogram is generated by counting the occurrence frequency of the local binary feature value of each original pixel within the color segmentation region. Each item in the mesh texture histogram represents the occurrence frequency of the corresponding local binary feature value. The preset uniform weave structure reference histogram is set according to the local binary feature values ​​of the corresponding region of the uniform weave structure. First, the region corresponding to the uniform weave structure is selected in the original underwater image. The local binary feature value of each original pixel in the region of the uniform weave structure is read. Then, the number of times the same local binary feature value appears is counted and arranged in ascending order of local binary feature value to obtain the preset uniform weave structure reference histogram. The preset morphological difference threshold is set according to the histogram difference between regions corresponding to uniformly woven structures and the histogram difference between regions corresponding to irregular shapes and regions corresponding to uniformly woven structures. First, the histogram difference between multiple regions corresponding to uniformly woven structures is calculated, and the maximum value of the histogram difference between multiple regions corresponding to uniformly woven structures is recorded as the first difference. Then, the histogram difference between regions corresponding to irregular shapes and regions corresponding to uniformly woven structures is calculated, and the minimum value of the histogram difference between regions corresponding to irregular shapes and regions corresponding to uniformly woven structures is recorded as the second difference. When the second difference is greater than the first difference, the preset morphological difference threshold = (first difference + second difference) / 2. When the second difference is less than or equal to the first difference, all histogram differences are sorted from low to high, and the median value of each two adjacent histogram differences is taken as the candidate morphological difference threshold. The number of regions corresponding to uniformly woven structures that are judged as irregular shapes and the number of regions corresponding to irregular shapes that are judged as non-irregular shapes are counted under the candidate morphological difference threshold. The candidate morphological difference threshold with the smallest sum of the two is used as the preset morphological difference threshold. The difference is the cumulative absolute value of the difference between the occurrence times of the mesh texture histogram and the preset uniform weave structure reference histogram at the same local binary feature value position. When the difference is greater than the preset shape difference threshold, the corresponding original pixel area is determined to be an irregular shape area. When the difference is equal to or less than the preset shape difference threshold, the corresponding original pixel area is not determined to be an irregular shape area.All irregularly shaped regions are extracted and mapped to the spatial coordinates of the original underwater image to generate a biofilm layer.

[0016] S3: Using a binary mask of scattered light spots and a water background layer, eliminate water scattering and fogging blurring interference in the original underwater image to generate an optimized underwater image, including: Extract the brightness aberration pixels marked with a value of one from the binary mask of scattered light spots. Divide the two-dimensional spatial interpolation window by the brightness aberration pixels. Calculate the ratio of the sum of the products of the pixel intensity of the adjacent original pixels marked with a value of zero and the corresponding spatial distance weights to the sum of the spatial distance weights to obtain the pixel intensity interpolation. Replace the pixel intensity of the brightness aberration pixels with the pixel intensity interpolation and rearrange them to generate a spotless image. Pixels with abnormal brightness marked as value one in the binary mask of scattered light spots are extracted. A two-dimensional spatial interpolation window is then defined based on the coordinate position of these pixels. This window represents the local coordinate range of adjacent original pixels surrounding the pixel with the abnormal brightness. The size of the two-dimensional spatial interpolation window is set according to the continuous distribution range of the pixels with abnormal brightness marked as value one in the binary mask. This continuous distribution range is obtained by statistically analyzing the connected regions formed by adjacent pixels with abnormal brightness. The pixel intensity of adjacent original pixels marked as value zero within the spatial interpolation window is read, along with the spatial distance between the pixel with the abnormal brightness and the adjacent original pixels marked as value zero. The spatial distance weight is set based on the spatial distance: the smaller the spatial distance, the larger the weight; conversely, the larger the spatial distance, the smaller the weight. The interpolation value of the pixel intensity is calculated as: (pixel intensity of adjacent original pixels marked as value zero within the spatial interpolation window × sum of their corresponding spatial distance weights) / sum of spatial distance weights. When there are adjacent original pixels marked with a value of zero within the spatial interpolation window, the interpolated pixel intensity is used as the replacement pixel intensity for the current pixel with abnormal brightness. When there are no adjacent original pixels marked with a value of zero within the spatial interpolation window, the two-dimensional spatial interpolation window is expanded and the image is reread. If there are still no adjacent original pixels marked with a value of zero within the expanded two-dimensional spatial interpolation window, the original pixel intensity of the pixel with abnormal brightness is used as the replacement pixel intensity for the current pixel with abnormal brightness. The replaced pixel intensities are then rearranged according to the spatial coordinates of the original underwater image to generate a spotless image.

[0017] In the water background layer, a background texture determination neighborhood is divided for each water background pixel. The pixel intensity gradient magnitude and pixel intensity variance of the water background pixels in each background texture determination neighborhood are calculated. When the pixel intensity gradient magnitude is less than a preset background gradient threshold and the pixel intensity variance is less than a preset background variance threshold, the corresponding background texture determination neighborhood is determined as a non-texture region. The average intensity of all water background pixels in the non-texture region in the water background layer is calculated to obtain the ambient light intensity of the water. In the water background layer, a background texture determination neighborhood is defined for each water background pixel. This neighborhood encompasses the range around the water background pixel used to determine local pixel intensity variations. The size of the background texture determination neighborhood is set based on the range of pixel intensity variations between adjacent water background pixels in the water background layer; the size of the neighborhood that can cover local pixel intensity variations is used as the background texture determination neighborhood size. The pixel intensity gradient magnitude is obtained from the pixel intensity difference between adjacent water background pixels within the background texture determination neighborhood, and the pixel intensity variance is obtained from the dispersion of the pixel intensity of the water background pixels within the background texture determination neighborhood relative to the average intensity. The preset background gradient threshold is set according to the distribution of pixel intensity gradient magnitudes in the water background layer. First, the pixel intensity gradient magnitude of each water background pixel within the neighborhood of each background texture determination is calculated. Then, all pixel intensity gradient magnitudes are arranged from low to high, and the 25th percentile pixel intensity gradient magnitude is taken as the preset background gradient threshold. The 25th percentile represents the portion of the water background layer where one-quarter of all pixel intensity gradient magnitudes are not greater than this value; the background texture determination neighborhood below this value corresponds to a gently changing portion of the water background layer. The preset background variance threshold is set according to the distribution of pixel intensity variance in the water background layer. First, the pixel intensity variance of each water background pixel within the neighborhood of each background texture determination is calculated. Then, all pixel intensity variances are arranged from low to high, and the 25th percentile pixel intensity variance is taken as the preset background variance threshold. The 25th percentile represents the portion of the water background layer where one-quarter of all pixel intensity variances are not greater than this value; the background texture determination neighborhood below this value corresponds to a portion of the water background layer where brightness dispersion is limited. When the pixel intensity gradient magnitude is less than a preset background gradient threshold and the pixel intensity variance is less than a preset background variance threshold, the corresponding background texture determination neighborhood is determined as a non-texture region; when the pixel intensity gradient magnitude is greater than or equal to the preset background gradient threshold, or the pixel intensity variance is greater than or equal to the preset background variance threshold, the corresponding background texture determination neighborhood is not determined as a non-texture region. The ambient light intensity of the water body is calculated as the sum of pixel intensities of all water background pixels in the non-texture region of the water background layer divided by the number of water background pixels in the non-texture region of the water background layer.

[0018] Extract the lowest intensity value of each pixel in the red, green and blue color channels of the spotless image, and find the minimum value of the lowest intensity value in the local neighborhood of the spotless image pixel to obtain the dark channel value of each spotless image pixel. Calculate the ratio of the dark channel value to the ambient light intensity of the water body, and subtract the product of the preset defogging coefficient and the ratio from a constant to obtain the water body transmittance. When extracting the lowest intensity value of each pixel in the red, green, and blue color channels of a spotless image, the color intensity of each pixel in the red, green, and blue color channels is read one by one, and the minimum color intensity among the three color channels is taken as the lowest intensity value. The local neighborhood of a pixel in a spotless image is the local range around which the minimum intensity value is calculated. The size of the local neighborhood is set according to the spatial range of the local mesh structure in the spotless image; the size within the same local mesh structure range is used as the local neighborhood size. The dark channel value is equal to the minimum of the lowest intensity values ​​within the local neighborhood of the pixel in the spotless image. The preset minimum water transmittance is set according to the ratio distribution between the dark channel value and the ambient light intensity of the water body. First, the ratio of the dark channel value to the ambient light intensity of the water body for each pixel in the spotless image is calculated. Then, all ratios are arranged from low to high, and the 90th percentile ratio is taken as the transmittance control ratio. The preset minimum water transmittance is 0.1, indicating that the water transmittance is retained as a positive number and not equal to 0. The preset dehazing coefficient = (1 - preset minimum water transmittance) / transmittance control ratio. When the transmittance control ratio is 0, the preset dehazing coefficient is set to 0. The ratio of the dark channel value to the ambient light intensity is calculated normally when the ambient light intensity is greater than zero, set to zero when the ambient light intensity is equal to zero, and recalculated when the ambient light intensity is less than zero. Water transmittance = 1 - preset defogging coefficient × ratio of dark channel value to ambient light intensity of water body; when water transmittance is greater than 1, it is corrected to 1; when it is equal to 1, it is retained as 1; when it is greater than 0 and less than 1, the water transmittance is retained; when it is less than or equal to 0, it is retained until subsequent steps are performed according to the preset reference water transmittance to obtain the water transmittance.

[0019] Extract the pixel intensity of each pixel in the spotless image as the observed intensity. Combine the observed intensity, the ambient light intensity of the water body, and the water transmittance to calculate the scene radiation of each pixel. The formula for calculating scene radiation is: In the formula, Indicates scene radiation. Indicates the intensity of observation. Indicates the ambient light intensity of the water body. This represents the water transmittance. When the water transmittance is less than or equal to zero, the current water transmittance is corrected to the preset reference water transmittance. The scene radiation at the coordinate position corresponding to each pixel in the spotless image is used as the descattering pixel intensity at the corresponding coordinate position. According to the coordinate position of the pixel in the spotless image in the original underwater image, the intensity of each descattering pixel is mapped to the pixel intensity at the corresponding coordinate position to replace the observation intensity at the corresponding coordinate position, thereby eliminating the interference of water scattering and fogging in the original underwater image and generating an optimized underwater image. The pixel intensity of each pixel in the spotless image is extracted and used as the observed intensity. Simultaneously, the ambient light intensity of the water body and the water transmittance at the same coordinate position are read. A preset reference water transmittance is set based on the valid values ​​of water transmittance greater than 0. First, all water transmittance values ​​greater than 0 are extracted. Then, the extracted water transmittance values ​​are arranged from low to high, and the 5th percentile of the sorted values ​​is taken as the preset reference water transmittance. The 5th percentile indicates that one-twentieth of all water transmittance values ​​greater than 0 are not greater than this value, which serves as the lower limit for positive values ​​when correcting water transmittance. If no water transmittance greater than 0 exists, the preset reference water transmittance is set to 0.1, meaning that the corrected water transmittance is still positive and not equal to 0. When the water transmittance is less than or equal to 0, the current water transmittance is corrected to the preset reference water transmittance; when the water transmittance is greater than 0 and less than or equal to 1, the current water transmittance is retained; when the water transmittance is greater than 1, the current water transmittance is corrected to 1. Scene radiation = (observation intensity - ambient light intensity) / water transmittance + ambient light intensity; when the scene radiation is less than the effective lower limit of pixel intensity, the effective lower limit of pixel intensity is used as the current scene radiation; when it is between the effective lower limit and the effective upper limit of pixel intensity, the original scene radiation is used as the current scene radiation; when it is greater than the effective upper limit of pixel intensity, the effective upper limit of pixel intensity is used as the current scene radiation. The scene radiation at the coordinate position corresponding to each pixel in the spotless image is used as the descattering pixel intensity at the corresponding coordinate position. According to the coordinate position of the pixel in the spotless image in the original underwater image, each descattering pixel intensity is mapped to the pixel intensity at the corresponding coordinate position, and the observation intensity at the corresponding coordinate position is replaced to generate the optimized underwater image.

[0020] S4: Referring to the biofouling layer, the netting structure layer, and the optimized underwater image, reconstruct a clear structural morphology of the netting without biofouling and water disturbance, generating a reconstructed netting image including: In the bio-attachment layer, a preset pixel scanning window is divided with each bio-attachment layer pixel as the center. The scanning window moves point by point on the bio-attachment layer according to its size. When the area covered by the pixel scanning window contains bio-attachment layer pixels in non-irregular shaped areas, or contains bio-attachment layer pixels in both non-irregular shaped areas and irregular shaped areas, the value of the bio-attachment layer pixel is set to zero. When the area covered by the pixel scanning window contains only bio-attachment layer pixels in irregular shaped areas, the value of the bio-attachment layer pixel is set to one, thus generating a binary mask for the bio-attachment region. Within the biofilm layer, a pre-defined pixel scanning window is defined, centered on each biofilm pixel. A biofilm pixel represents the pixel position within the biofilm layer corresponding to the spatial coordinates of the original underwater image. The pixel scanning window represents a local area on the biofilm layer that is moved point-by-point to read region attributes. The pre-defined pixel scanning window is set according to the continuous distribution range of irregularly shaped regions within the biofilm layer. First, the number of biofilm pixels in adjacent irregularly shaped regions is counted. Then, the maximum span of adjacent irregularly shaped regions in the horizontal and vertical coordinate directions is calculated. The smaller of the two maximum spans is taken as the candidate window side length. If the candidate window side length is odd, it is used as the pre-defined pixel scanning window side length; if it is even, it is incremented by 1 and used as the pre-defined pixel scanning window side length. The scanning window moves point-by-point across the biofilm layer according to its size, and each movement reads the region attributes corresponding to the biofilm pixels within the area covered by the pixel scanning window. If the area covered by the pixel scanning window contains bio-attachment layer pixels from non-irregular regions, or contains bio-attachment layer pixels from both non-irregular and irregular regions, the value of the bio-attachment layer pixels is set to zero. If the area covered by the pixel scanning window contains only bio-attachment layer pixels from irregular regions, the value of the bio-attachment layer pixels is set to one. If there are no valid bio-attachment layer pixels in the area covered by the pixel scanning window, the value of the bio-attachment layer pixels is set to zero, thus generating a binary mask for the bio-attachment region.

[0021] Extract the corresponding coordinate values ​​of the pixels in the bio-attachment layer of the binary mask of the bio-attachment region as the bio-attachment mask values. Simultaneously, construct a matrix of the same size as the binary mask of the bio-attachment region, consisting entirely of values ​​of 1. Multiply the mesh structure feature values ​​in the mesh structure layer by the preset maximum pixel intensity and perform inverse normalization. Combine the scene radiation and the bio-attachment mask values ​​to calculate the reconstructed pixel values. The formula for calculating the reconstructed pixel values ​​is as follows: In the formula, Indicates the coordinate position in the reconstructed mesh image The reconstructed pixel values, Indicates the coordinate position in the mesh structure layer The eigenvalues ​​of the mesh structure after inverse normalization are the numerical values. Indicates the coordinate position in the optimized underwater image. Scene radiation, Indicates the coordinate position of the bio-attachment area mask. The biofilm mask value is used, where the number 1 represents the value at the corresponding coordinate position in a matrix where all values ​​are 1; the reconstructed pixel values ​​are arranged and combined to reconstruct a clear structure of the net without biofilm or water interference, generating a reconstructed net image. Extract the corresponding values ​​of the coordinate positions of the pixels in the bioattachment layer of the binary mask of the bioattachment region, and use these values ​​as the bioattachment mask values. The bioattachment mask values ​​are the values ​​at corresponding coordinate positions in the binary mask of the bioattachment region; a value of 1 corresponds to an irregularly shaped region, and a value of 0 corresponds to a non-irregularly shaped region or a region containing both irregularly shaped and non-irregularly shaped regions. Construct a matrix of the same size as the binary mask of the bioattachment region, containing only values ​​of 1. Each coordinate position in this matrix corresponds to a coordinate position in the binary mask of the bioattachment region. The preset maximum pixel intensity is set according to the pixel intensity range of the original underwater image. First, read the pixel intensity of all original pixels in the original underwater image and determine the maximum pixel intensity; this maximum pixel intensity is used as the preset maximum pixel intensity. The inversely normalized value of the mesh structure feature value at the coordinate position in the mesh structure layer is equal to the mesh structure feature value at the coordinate position in the mesh structure layer multiplied by the preset maximum pixel intensity. When the bio-attachment mask value is one, the inversely normalized value of the mesh structure feature at the coordinate position in the mesh structure layer is used as the current reconstructed pixel value. When the bio-attachment mask value is zero, the scene radiation at the coordinate position in the optimized underwater image is used as the current reconstructed pixel value. When the bio-attachment mask value is greater than zero and less than one, the reconstructed pixel value = bio-attachment mask value × inversely normalized value of the mesh structure feature at the coordinate position in the mesh structure layer + (1 - bio-attachment mask value) × scene radiation at the coordinate position in the optimized underwater image. The reconstructed mesh image is generated by arranging and combining the reconstructed pixel values.

[0022] S5: Determine the set of pixels representing continuous lines of the mesh in the reconstructed mesh image, and calculate the gap length of the mesh fibers based on the set of pixels representing continuous lines of the mesh, including: Calculate the intensity gradient magnitude of each reconstructed mesh pixel in the reconstructed mesh image. If the intensity gradient magnitude of the current reconstructed mesh pixel is less than the intensity gradient magnitude of any adjacent reconstructed mesh pixel, then set the intensity gradient magnitude of the current reconstructed mesh pixel to zero, and retain the reconstructed mesh pixels with intensity gradient magnitude greater than zero as candidate edge pixels. When calculating the intensity gradient magnitude of each reconstructed mesh pixel in the reconstructed mesh image, the pixel intensity of the reconstructed mesh pixel is read, along with the pixel intensities of its adjacent reconstructed mesh pixels. A reconstructed mesh pixel is a pixel in the reconstructed mesh image that has both coordinate position and pixel intensity. Adjacent reconstructed mesh pixels are pixels surrounding the reconstructed mesh pixel that participate in the intensity gradient magnitude comparison. The intensity gradient magnitude equals the change in intensity caused by the difference in pixel intensity between the reconstructed mesh pixel and its adjacent reconstructed mesh pixels. If the intensity gradient magnitude of the current reconstructed mesh pixel is less than the intensity gradient magnitude of any adjacent reconstructed mesh pixel, the intensity gradient magnitude of the current reconstructed mesh pixel is set to zero. If it is equal to the intensity gradient magnitude of an adjacent reconstructed mesh pixel and not less than the intensity gradient magnitudes of other adjacent reconstructed mesh pixels, or greater than the intensity gradient magnitudes of all adjacent reconstructed mesh pixels, the intensity gradient magnitude of the current reconstructed mesh pixel is retained. Reconstructed mesh pixels with an intensity gradient magnitude greater than zero are considered candidate edge pixels; reconstructed mesh pixels with an intensity gradient magnitude equal to zero are not considered candidate edge pixels; when the intensity gradient magnitude is less than zero, the intensity gradient magnitude is recalculated based on the pixel intensity of adjacent reconstructed mesh pixels to obtain candidate edge pixels.

[0023] Candidate edge pixels with intensity gradient magnitude greater than a preset strong edge threshold are marked as strong edge pixels. Candidate edge pixels with intensity gradient magnitude between a preset weak edge threshold and a preset strong edge threshold are marked as weak edge pixels. Candidate edge pixels with intensity gradient magnitude less than a preset weak edge threshold are removed. It is determined whether there is a strong edge pixel in the eight-neighbor range of each weak edge pixel. If there is, the corresponding weak edge pixel is retained and converted into a strong edge pixel. If there is no strong edge pixel, the corresponding weak edge pixel is removed. Adjacent strong edge pixels in the eight-neighbor range are connected in sequence, and the connected strong edge pixels are used as line pixels to generate a set of continuous line pixels for the mesh. The intensity gradient magnitude of candidate edge pixels is read, and preset strong edge thresholds and preset weak edge thresholds are retrieved. The preset weak edge thresholds and preset strong edge thresholds are set according to the distribution of intensity gradient magnitudes of candidate edge pixels. First, the intensity gradient magnitudes of all candidate edge pixels are arranged from low to high. Then, the intensity gradient magnitude at the 30th percentile of the arrangement is taken as the preset weak edge threshold, and the intensity gradient magnitude at the 70th percentile is taken as the preset strong edge threshold. The 30th percentile indicates that three-tenths of the intensity gradient magnitudes of all candidate edge pixels are not greater than this value, which is used to limit the lower bound for entering the weak edge point judgment. The 70th percentile indicates that seven-tenths of the intensity gradient magnitudes of all candidate edge pixels are not greater than this value, which is used to limit the lower bound for entering the strong edge point judgment. When the preset strong edge threshold is less than the preset weak edge threshold, the preset strong edge threshold and the preset weak edge threshold are swapped. When the preset strong edge threshold is equal to the preset weak edge threshold, candidate edge pixels with the same threshold are marked as weak edge points. When the intensity gradient magnitude is greater than a preset strong edge threshold, the candidate edge pixel is marked as a strong edge pixel; when the intensity gradient magnitude is greater than or equal to a preset weak edge threshold and less than or equal to a preset strong edge threshold, the candidate edge pixel is marked as a weak edge pixel; when the intensity gradient magnitude is less than a preset weak edge threshold, the candidate edge pixel is discarded. If a strong edge pixel exists within an eight-neighborhood, the corresponding weak edge pixel is retained and converted into a strong edge pixel; if no strong edge pixel exists within an eight-neighborhood, the corresponding weak edge pixel is discarded. Adjacent strong edge pixels within the eight-neighborhood are connected sequentially, and the connected strong edge pixels are used as line pixels to generate a continuous set of grid line pixels.

[0024] The number of adjacent line pixels in the eight-neighbor range of each line pixel in the continuous line pixel set of the mesh is counted. The line pixels with one adjacent line pixel are taken as the endpoints of the discontinuous gaps in the mesh. The straight-line distance between the two discontinuous gap endpoints of the same mesh is calculated to obtain the gap length of the mesh fiber. When counting the number of neighboring line pixels within the eight-neighborhood of each line pixel in a continuous halftone pixel set, the coordinates of the line pixel are read, and neighboring line pixels within the eight-neighborhood belonging to the continuous halftone pixel set are retrieved. If the number of neighboring line pixels is one, the line pixel is considered a discontinuous endpoint of the halftone line; if the number of neighboring line pixels is zero, two, or greater than two, the line pixel is not considered a discontinuous endpoint. A discontinuous endpoint is the end pixel in the continuous halftone pixel set that is connected to only one neighboring line pixel. The same halftone line is determined based on the continuous connection relationship between line pixels in the continuous halftone pixel set, and is traced point by point along the adjacent line pixels to the discontinuous endpoint. The straight-line distance between two discontinuous endpoints equals the spatial distance between their coordinates. When the same network cable corresponds to two discontinuous empty endpoints, the straight-line distance between the two discontinuous empty endpoints is taken as the current gap length of the network cable fiber; when the same network cable corresponds to fewer than two discontinuous empty endpoints, the gap length of the network cable fiber is not calculated; when the same network cable corresponds to more than two discontinuous empty endpoints, the straight-line distance between two adjacent discontinuous empty endpoints is taken as the current gap length of the network cable fiber according to the connection order in the pixel set of the continuous lines of the network cable, and the gap length of the network cable fiber is obtained.

[0025] S6: Scan the pixel set of continuous lines of the mesh, calculate the percentage reduction in the width of the mesh fibers and the texture feature deviation of the mesh fibers, and combine this with the gap length of the mesh fibers to determine whether there are gaps, width reduction, and texture feature deviation in the mesh fibers, generating the mesh clothing wear image recognition results, including: Scan the set of continuous line pixels of the network and determine whether the removal of the outermost line pixel in the set of continuous line pixels will destroy the overall connectivity of the set of continuous line pixels. Remove the outermost line pixels that do not destroy connectivity and are removed layer by layer until the width of the remaining line pixel area becomes a single pixel. Use the remaining single pixel width area as a single pixel wide fiber skeleton. When scanning a continuous set of line pixels, the coordinates of each line pixel are read, and it is determined whether the line pixel is located on the outermost edge. If there are non-line pixels within the eight-neighborhood, the line pixel is considered the outermost line pixel; if all eight-neighborhoods are line pixels, the line pixel is not considered the outermost line pixel. To determine whether removing the outermost line pixel in the continuous set of line pixels would disrupt the overall connectivity of the set, the outermost line pixel is temporarily removed, and the remaining line pixels are checked to see if they are still continuously connected through their eight-neighborhood relationships. Line pixels that do not disrupt connectivity after removal and are located on the outermost edge are eliminated layer by layer; line pixels that disrupt connectivity after removal are retained. While the line pixel region width has not yet become a single pixel, new outermost line pixels are scanned; when the line pixel region width becomes a single pixel, the layer-by-layer elimination stops. The remaining single-pixel-width region is used as a single-pixel-width fiber skeleton.

[0026] Calculate the minimum straight-line distance from each line pixel on the single-pixel wide fiber skeleton to the outermost line pixel closest to that line pixel, estimate the local width of the mesh fiber, and at the same time calculate the ratio of the difference between the preset healthy fiber width and the local width of the mesh fiber to the preset healthy fiber width to obtain the percentage reduction in the width of the mesh fiber. When calculating the minimum straight-line distance from each line pixel on the single-pixel wide fiber skeleton to the nearest outermost line pixel, the coordinates of the line pixels on the single-pixel wide fiber skeleton are read, along with the coordinates of the outermost line pixel in the continuous line pixel set of the mesh. The minimum straight-line distance is the minimum straight-line distance between each line pixel on the single-pixel wide fiber skeleton and each outermost line pixel. The local width of the mesh fiber is estimated based on this minimum straight-line distance. The preset healthy fiber width is set according to the local width distribution of the mesh fibers. First, continuous mesh regions without discontinuous endpoints are extracted from the continuous line pixel set. Then, the local width of the mesh fiber corresponding to each line pixel on the single-pixel wide fiber skeleton within the continuous mesh region is calculated. All local widths of the mesh fibers are arranged from low to high values, and the local width of the mesh fiber at the 95th percentile after arrangement is taken as the preset healthy fiber width. The 95th percentile indicates that 95% of the mesh fibers within the continuous mesh region have a local width no greater than this value, which serves as the upper limit of the healthy fiber width. The percentage reduction in network cable fiber width is calculated as follows: (Preset healthy fiber width - Local width of network cable fiber) / Preset healthy fiber width. When the local width of the network cable fiber is less than the preset healthy fiber width, the current percentage reduction in network cable fiber width is calculated using the formula. When the local width of the network cable fiber is equal to or greater than the preset healthy fiber width, the value is zero, which is taken as the current percentage reduction in network cable fiber width.

[0027] In the reconstructed mesh image, the brightness value of each line pixel in the continuous line pixel set of the mesh is compared with the brightness value of the corresponding surrounding line pixels. The surrounding line pixels with a brightness value greater than or equal to the brightness value of the line pixel are marked as one, and the surrounding line pixels with a brightness value less than the brightness value of the line pixel are marked as zero. The marking results are combined into binary numbers in clockwise order of eight neighborhoods and converted into corresponding decimal texture values. The resulting array is arranged into a mesh fiber texture feature map. The sum of the absolute differences between the decimal texture values ​​at the same line pixel coordinate position in the preset healthy mesh fiber reference texture feature map and the mesh fiber texture feature map is calculated to obtain the texture feature deviation of the mesh fiber. In the reconstructed mesh image, the brightness value of each pixel in the continuous mesh line pixel set is read, along with the brightness values ​​of the corresponding surrounding pixels. The surrounding pixels are those located around the original pixel in the continuous mesh line pixel set. A pixel with a brightness value greater than or equal to the original pixel's brightness value is marked as 1, and one with a brightness value less than the original pixel's brightness value is marked as 0. The marking results are combined into binary numbers in a clockwise eight-neighbor order and converted into corresponding decimal texture values. These decimal texture values ​​are then arranged according to the pixel coordinates to form the mesh fiber texture feature map. A preset healthy mesh fiber reference texture feature map is created based on pixels in the continuous mesh line pixel set that do not have discontinuous mesh line endpoints or are not classified as irregularly shaped regions. The brightness values ​​of these pixels and their corresponding surrounding pixels are read, and decimal texture values ​​are generated in the same clockwise eight-neighbor order. Finally, these decimal texture values ​​are arranged according to the pixel coordinates to form the preset healthy mesh fiber reference texture feature map. When the coordinates of identical line pixels exist in both the preset healthy network fiber reference texture feature map and the network fiber texture feature map, the absolute difference is equal to the absolute value of the difference between the decimal texture values ​​in the preset healthy network fiber reference texture feature map and the decimal texture values ​​in the network fiber texture feature map. If the coordinates of identical line pixels exist only in the preset healthy network fiber reference texture feature map or only in the network fiber texture feature map, they are not included in the accumulation. The texture feature deviation of the network fiber is the sum of all accumulated absolute differences.

[0028] The system summarizes the gap length of the mesh fibers, the percentage reduction in the width of the mesh fibers, and the deviation in the texture features of the mesh fibers to determine whether there are gaps, width reduction, and texture feature deviations in the mesh fibers, and generates the image recognition results of the mesh wear of the snagging mesh. When summarizing the gap length, width reduction percentage, and texture feature deviation of network cables, these parameters are mapped to the same network cable. The preset gap length threshold is set according to the gap length distribution of network cables. First, the gap lengths of all network cables with a value greater than 0 are extracted, and the gap lengths are arranged from low to high. The gap length of the 75th percentile of the arranged network cables is taken as the preset gap length threshold. If no network cable has a gap length greater than 0, the threshold judgment for the existence of a gap is not performed, and the network cable is determined to have no gap. The preset width reduction percentage threshold is set according to the width reduction percentage distribution of network cable fibers. First, the width reduction percentage of all network cable fibers with a value greater than 0 is extracted. These percentages are then arranged from lowest to highest value, and the 75th percentile of the sorted fiber's width reduction percentage is taken as the preset width reduction percentage threshold. If no network cable fiber has a width reduction percentage greater than 0, the threshold judgment for whether the fiber has a width reduction is not performed, and the fiber is determined to have no width reduction. The preset texture feature deviation threshold is set according to the texture feature deviation distribution of network cable fibers. First, the texture feature deviation of all network cable fibers with a value greater than 0 is extracted. These deviations are then arranged from lowest to highest value, and the 75th percentile of the sorted fiber's texture feature deviation is taken as the preset texture feature deviation threshold. If no network cable fiber has a texture feature deviation greater than 0, the threshold judgment for whether the fiber has a texture feature deviation is not performed, and the fiber is determined to have no texture feature deviation. The 75th percentile indicates that three-quarters of the corresponding indicator values ​​are not greater than this value; if the corresponding indicator exceeds this value, the existence status judgment is performed. If a gap length greater than 0 exists in the network cable fiber and this gap length is greater than a preset gap length threshold, the network cable fiber is determined to have a gap. If a gap length greater than 0 exists in the network cable fiber and this gap length is equal to or less than the preset gap length threshold, the network cable fiber is determined not to have a gap. If a width reduction percentage greater than 0 exists in the network cable fiber and this width reduction percentage is greater than a preset width reduction percentage threshold, the network cable fiber is determined to have a width reduction. If a width reduction percentage greater than 0 exists in the network cable fiber and this width reduction percentage is equal to or less than the preset width reduction percentage threshold, the network cable fiber is determined not to have a width reduction. If a texture feature deviation greater than 0 exists in the network cable fiber and this texture feature deviation is greater than a preset texture feature deviation threshold, the network cable fiber is determined to have a texture feature deviation. If a texture feature deviation greater than 0 exists in the network cable fiber and this texture feature deviation is equal to or less than the preset texture feature deviation threshold, the network cable fiber is determined not to have a texture feature deviation.When there are no gaps in the mesh fibers, no reduction in width, and no deviation in texture features, an image recognition result for the wear of the sifting netting is generated that does not identify wear. When there are any of the following conditions in the mesh fibers: gaps, reduction in width, and deviation in texture features, an image recognition result for the wear of the sifting netting is generated that identifies wear.

[0029] Please see Figure 2 A system for recognizing the wear of underwater sorting nets is provided for identifying the wear condition of the nets during underwater sorting net operations. The system includes: The image layering module acquires raw underwater images of the sifting net during its underwater operation period and determines the net structure layer and water background layer in the raw underwater images. The interference feature extraction module determines the binary mask of scattered light spots in the original underwater image, and converts the original underwater image from RGB color space to HSV color space to determine the biological attachment layer in the original underwater image. The background optimization module, referencing the binary mask of scattered light spots and the water background layer, eliminates water scattering and fogging blur interference in the original underwater image and generates an optimized underwater image. The image reconstruction module, referring to the bio-attached layer, the net structure layer and the optimized underwater image, reconstructs a clear structural morphology of the net without bio-attachment and water disturbance, and generates a reconstructed net image. The line feature extraction module determines the set of pixels of continuous lines in the reconstructed mesh image and calculates the gap length of the mesh fibers based on the set of pixels of continuous lines. The mesh wear recognition module scans the pixel set of continuous lines of the mesh, calculates the percentage reduction in the width of the mesh fibers and the deviation in the texture features of the mesh fibers, and combines the gap length of the mesh fibers to determine whether there are gaps, width reduction and texture feature deviation of the mesh fibers, and generates the mesh wear image recognition result of the sifting net.

[0030] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for image recognition of abrasion on underwater sorting nets, used to identify the abrasion condition of the net during underwater sorting net operations, characterized in that... Includes the following steps: Collect raw underwater images of the sifting net during its underwater operation period and determine the net structure layer and water background layer in the raw underwater images; The binary mask of scattered light spots in the original underwater image is determined, and the original underwater image is converted from RGB color space to HSV color space to determine the bio-attachment layer in the original underwater image. By referring to the binary mask of scattered light spots and the water background layer, the interference of water scattering and fogging in the original underwater image is eliminated, and the optimized underwater image is generated. By referring to the bio-attached layer, the net structure layer and the optimized underwater image, the clear structural morphology of the net without bio-attachment and water disturbance is reconstructed, and the reconstructed net image is generated. Determine the set of pixels of continuous lines of mesh in the reconstructed mesh image, and calculate the gap length of mesh fibers based on the set of pixels of continuous lines of mesh; Scan the pixel set of continuous lines of the mesh, calculate the percentage reduction in the width of the mesh fibers and the texture feature deviation of the mesh fibers, and combine the gap length of the mesh fibers to determine whether there are gaps, width reduction and texture feature deviation of the mesh fibers, and generate the image recognition result of mesh wear of the snagging mesh.

2. The method for image recognition of mesh wear on a sifting net according to claim 1, characterized in that, The acquisition of raw underwater images of the winch net during its underwater operation period, and the determination of the net structure layer and water background layer in the raw underwater images, include: Collect raw underwater images of the sorting net during the operation period, calculate the pixel intensity gradient value of each raw pixel in the raw underwater image with its horizontal and vertical adjacent raw pixels, and calculate the gradient magnitude of each raw pixel based on the horizontal and vertical pixel intensity gradient values. Each original pixel in the original underwater image is regarded as the central original pixel. The pixel neighborhood window is divided, and the variance of the mesh pixel intensity within each pixel neighborhood window is calculated. The gradient magnitude of each original pixel and the variance of the mesh pixel intensity are normalized respectively. The weighted sum of the normalized gradient magnitude and the normalized mesh pixel intensity variance is calculated to obtain the mesh structure feature value. The feature values ​​of the mesh structure corresponding to each original pixel are arranged according to the spatial coordinate distribution of the original underwater image to generate a mesh structure layer. At the same time, the pixel intensity of the central original pixel and the corresponding surrounding original pixels in the original underwater image are calculated by weighted average. Based on the calculation results, the edge details in the original underwater image are smoothed and the brightness distribution information of the water body is preserved to generate a water background layer.

3. The image recognition method for wear and tear of the sifting netting according to claim 1, characterized in that, The process of determining the scattered light spot binary mask in the original underwater image, converting the original underwater image from RGB color space to HSV color space, and determining the biofilm layer in the original underwater image includes: For each central original pixel in the original underwater image, an analysis neighborhood is divided. The peak pixel intensity in each analysis neighborhood is detected. The difference between the peak pixel intensity and the preset global brightness threshold is calculated to filter out pixels with abnormal brightness in the water body. At the same time, a binary mask matrix with the same size as the original underwater image is constructed. The mask value of the coordinate position of the pixel with abnormal brightness in the binary mask matrix is ​​marked as the value one, and the mask value of the coordinate position of the original pixel that is not identified as a pixel with abnormal brightness is marked as the value zero. The marked binary mask matrix is ​​used as a binary mask for scattered light spots. The original underwater image is converted from RGB color space to HSV color space. Hue and saturation in HSV color space are extracted. When the hue is within the preset range of marine life hue and the saturation is greater than the preset saturation threshold, the original pixels corresponding to the hue and saturation are retained. When the hue exceeds the preset range of marine life hue or the saturation is less than or equal to the preset saturation threshold, the original pixels corresponding to the hue and saturation are removed to obtain the color segmentation region. The original underwater image is converted into a grayscale image. In the grayscale image, the grayscale value of each original pixel is compared with that of its corresponding neighboring original pixels. If the grayscale value of the neighboring original pixel is greater than or equal to that of the original pixel, it is recorded as one; otherwise, it is recorded as zero. The values ​​corresponding to each neighboring original pixel are combined into a binary sequence and converted into decimal local binary feature values. The local binary feature values ​​of each original pixel within the color segmentation region are statistically analyzed to generate a corresponding mesh texture histogram. The difference between the mesh texture histogram and the preset uniform weave structure reference histogram is calculated. When the difference is greater than a preset morphological difference threshold, the corresponding original pixel region is determined to be an irregular morphological region. All irregular morphological regions are extracted and mapped according to the spatial coordinates of the original underwater image to generate a bio-attachment layer.

4. The image recognition method for wear and tear of the sifting netting according to claim 3, characterized in that, The reference scattered light spot binary mask and water background layer eliminate water scattering and fogging blur interference in the original underwater image, generating an optimized underwater image including: Extract the brightness aberration pixels marked with a value of one from the binary mask of scattered light spots, divide the two-dimensional spatial interpolation window with the brightness aberration pixels, calculate the ratio of the sum of the product of the pixel intensity of the adjacent original pixels marked with a value of zero and the corresponding spatial distance weight to the sum of the spatial distance weights, and obtain the pixel intensity interpolation. Replace the pixel intensity of the brightness aberration pixels with the pixel intensity interpolation and rearrange them to generate a spotless image. In the water background layer, a background texture determination neighborhood is divided for each water background pixel. The pixel intensity gradient magnitude and pixel intensity variance of the water background pixels in each background texture determination neighborhood are calculated. When the pixel intensity gradient magnitude is less than a preset background gradient threshold and the pixel intensity variance is less than a preset background variance threshold, the corresponding background texture determination neighborhood is determined as a non-texture region. The average intensity of all water background pixels in the non-texture region of the water background layer is calculated to obtain the ambient light intensity of the water. Extract the lowest intensity value of each pixel in the red, green and blue color channels of the spotless image, and find the minimum value of the lowest intensity value in the local neighborhood of the spotless image pixel to obtain the dark channel value of each spotless image pixel. Calculate the ratio of the dark channel value to the ambient light intensity of the water body, and subtract the product of the preset defogging coefficient and the ratio from a constant to obtain the water body transmittance. The pixel intensity of each pixel in the spotless image is extracted as the observed intensity. The scene radiation of each pixel is calculated by combining the observed intensity, the ambient light intensity of the water body, and the water transmittance. The scene radiation at the corresponding coordinate position of each pixel is used as the descattering pixel intensity at the corresponding coordinate position. According to the coordinate position of the pixel in the original underwater image, each descattering pixel intensity is mapped to the pixel intensity at the corresponding coordinate position to replace the observed intensity at the corresponding coordinate position. This eliminates the interference of water scattering and fogging in the original underwater image and generates an optimized underwater image.

5. The image recognition method for wear on the mesh of a sifting net according to claim 4, characterized in that, The formula for calculating the scene radiation is: ； In the formula, Indicates scene radiation. Indicates the intensity of observation. Indicates the ambient light intensity of the water body. This represents the water body transmittance. When the water body transmittance is less than or equal to zero, the current water body transmittance is corrected to the preset reference water body transmittance.

6. The image recognition method for wear on the mesh of a sifting net according to claim 4, characterized in that, The process involves reconstructing a clear structural morphology of the netting without biofouling or water disturbance, based on the reference biofouling layer, the netting structure layer, and the optimized underwater image. The resulting reconstructed netting image includes: In the bio-attachment layer, a preset pixel scanning window is divided with each bio-attachment layer pixel as the center. The scanning window moves point by point on the bio-attachment layer according to its size. When the area covered by the pixel scanning window contains bio-attachment layer pixels in non-irregular shaped areas, or contains bio-attachment layer pixels in both non-irregular shaped areas and irregular shaped areas, the value of the bio-attachment layer pixel is set to zero. When the area covered by the pixel scanning window contains only bio-attachment layer pixels in irregular shaped areas, the value of the bio-attachment layer pixel is set to one, thus generating a binary mask for the bio-attachment region. The corresponding values ​​of the coordinate positions of the pixels in the binary mask of the bioattachment region are extracted and used as the bioattachment mask values. At the same time, a matrix with all values ​​of 1 and the same size as the binary mask of the bioattachment region is constructed. The feature values ​​of the mesh structure in the mesh structure layer are multiplied by the preset maximum pixel intensity and inversely normalized. Combined with the scene radiation and the bioattachment mask values, the reconstructed pixel values ​​are calculated and arranged to reconstruct a clear mesh structure without bioattachment and water interference, generating a reconstructed mesh image.

7. The method for image recognition of wear on the mesh fabric of a sifting net according to claim 6, characterized in that, The formula for calculating the reconstructed pixel value is: ； In the formula, Indicates the coordinate position in the reconstructed mesh image. The reconstructed pixel values, Indicates the coordinate position in the mesh structure layer The eigenvalues ​​of the mesh structure after inverse normalization are the numerical values. Indicates the coordinate position in the optimized underwater image. Scene radiation, Indicates the coordinate position of the bio-attachment area mask. The biological attachment mask value, where the number 1 represents the value at the corresponding coordinate position in a matrix where all values ​​are 1.

8. The method for image recognition of wear on the mesh fabric of a sifting net according to claim 6, characterized in that, The process of determining the set of pixels representing continuous lines of mesh in the reconstructed mesh image, and calculating the gap length of the mesh fibers based on the set of pixels representing continuous lines of mesh, includes: Calculate the intensity gradient magnitude of each reconstructed mesh pixel in the reconstructed mesh image. If the intensity gradient magnitude of the current reconstructed mesh pixel is less than the intensity gradient magnitude of any adjacent reconstructed mesh pixel, then set the intensity gradient magnitude of the current reconstructed mesh pixel to zero, and retain the reconstructed mesh pixels with intensity gradient magnitude greater than zero as candidate edge pixels. Candidate edge pixels with intensity gradient magnitude greater than a preset strong edge threshold are marked as strong edge pixels. Candidate edge pixels with intensity gradient magnitude between a preset weak edge threshold and a preset strong edge threshold are marked as weak edge pixels. Candidate edge pixels with intensity gradient magnitude less than a preset weak edge threshold are removed. It is determined whether there is a strong edge pixel in the eight-neighbor range of each weak edge pixel. If there is, the corresponding weak edge pixel is retained and converted into a strong edge pixel. If there is no strong edge pixel, the corresponding weak edge pixel is removed. Adjacent strong edge pixels in the eight-neighbor range are connected in sequence, and the connected strong edge pixels are used as line pixels to generate a set of continuous line pixels for the mesh. The number of adjacent line pixels in the eight-neighbor range of each line pixel in the continuous line pixel set of the network is counted. The line pixels with one adjacent line pixel are taken as the endpoints of the network discontinuity gap. The straight-line distance between the two endpoints of the network discontinuity gap corresponding to the same network is calculated to obtain the gap length of the network fiber.

9. The method for image recognition of mesh wear on a sifting net according to claim 8, characterized in that, The scanned continuous line pixel set is used to calculate the percentage reduction in the width of the mesh fibers and the texture feature deviation of the mesh fibers. Combined with the gap length of the mesh fibers, it is determined whether there are gaps, width reduction, and texture feature deviation in the mesh fibers, generating the mesh clothing wear image recognition result, including: Scan the set of continuous line pixels of the network cable, determine whether the removal of the outermost line pixel in the set of continuous line pixels of the network cable will destroy the overall connectivity of the set of continuous line pixels of the network cable, remove the outermost line pixels that do not destroy the connectivity after removal layer by layer until the width of the remaining line pixel area becomes a single pixel, and use the remaining single pixel width area as a single pixel wide fiber skeleton. Calculate the minimum straight-line distance from each line pixel on the single-pixel wide fiber skeleton to the outermost line pixel closest to that line pixel, estimate the local width of the network fiber, and simultaneously calculate the ratio of the difference between the preset healthy fiber width and the local width of the network fiber to the preset healthy fiber width to obtain the percentage reduction in the width of the network fiber. In the reconstructed mesh image, the brightness value of each line pixel in the continuous line pixel set of the mesh is compared with the brightness value of the corresponding surrounding line pixels. The surrounding line pixels with a brightness value greater than or equal to the brightness value of the line pixel are marked as one, and the surrounding line pixels with a brightness value less than the brightness value of the line pixel are marked as zero. The marking results are combined into binary numbers in clockwise order of eight neighborhoods and converted into corresponding decimal texture values. The resulting array is arranged into a mesh fiber texture feature map. The sum of the absolute differences between the decimal texture values ​​at the same line pixel coordinate position in the preset healthy mesh fiber reference texture feature map and the mesh fiber texture feature map is calculated to obtain the texture feature deviation of the mesh fiber. By summarizing the gap length of the mesh fibers, the percentage reduction in the width of the mesh fibers, and the texture feature deviation of the mesh fibers, it is determined whether there are gaps, width reduction, and texture feature deviations in the mesh fibers, and a wear image recognition result of the mesh fabric is generated.

10. A system for recognizing the wear of underwater sorting nets, used to identify the wear condition of the netting during underwater sorting net operations, characterized in that... The system includes: The image layering module acquires raw underwater images of the sifting net during its underwater operation period and determines the net structure layer and water background layer in the raw underwater images. The interference feature extraction module determines the binary mask of scattered light spots in the original underwater image, and converts the original underwater image from RGB color space to HSV color space to determine the biological attachment layer in the original underwater image. The background optimization module, referencing the binary mask of scattered light spots and the water background layer, eliminates water scattering and fogging blur interference in the original underwater image and generates an optimized underwater image. The image reconstruction module, referring to the bio-attached layer, the net structure layer and the optimized underwater image, reconstructs a clear structural morphology of the net without bio-attachment and water disturbance, and generates a reconstructed net image. The line feature extraction module determines the set of pixels of continuous lines in the reconstructed mesh image and calculates the gap length of the mesh fibers based on the set of pixels of continuous lines. The mesh wear recognition module scans the pixel set of continuous lines of the mesh, calculates the percentage reduction in the width of the mesh fibers and the deviation in the texture features of the mesh fibers, and combines the gap length of the mesh fibers to determine whether there are gaps, width reduction and texture feature deviation of the mesh fibers, and generates the mesh wear image recognition result of the sifting net.

Images