A computer vision and deep learning combined leather product feature recognition method

Abstract

The present application relates to the technical field of leather detection, and specifically discloses a leather product feature recognition method based on computer vision and deep learning, which comprises: obtaining visual detection information of the leather to be detected in a preset air puffing state, wherein the visual detection information comprises light transmission information and elastic deformation information; obtaining leather dark injury information according to the light transmission information, wherein the leather dark injury information comprises leather texture dark injury information and leather processing dark injury information. The present application synchronously collects the surface texture image and the backside transmission light image of the leather in the air puffing and stretching state, effectively fuses the surface visual features and the internal structure information. The transmission light imaging can penetrate the material, clearly present the hidden defects such as internal dark injury, uneven density and micro-cracks which are difficult to be found by traditional methods, overcome the detection blind area caused by reflection and texture interference of single surface imaging, and significantly improve the coverage range and accuracy of defect recognition.

Description

Technical Field

[0001] This invention relates to the field of leather inspection technology, and in particular to a method for feature recognition of leather products based on a combination of computer vision and deep learning. Background Technology

[0002] Leather feature identification technology plays a key role in luxury goods authentication, industrial quality inspection and supply chain management. Its core objective is to achieve automated assessment of leather types, defect distribution and quality grades.

[0003] Existing leather feature recognition methods rely solely on visible light imaging of the surface, failing to effectively acquire information about the internal structure of leather when photographing and analyzing leather in a static or natural state. Naturally occurring leather often exhibits significant specular reflection, high texture similarity, uneven dyeing, and aging discoloration, making surface images susceptible to interference from high light saturation and texture confusion. This makes it difficult to clearly reveal hidden defects such as internal blemishes, microcracks, and fiber breaks. Summary of the Invention

[0004] The purpose of this invention is to provide a feature recognition method for leather products based on the combination of computer vision and deep learning, so as to solve the technical problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for feature recognition of leather products based on a combination of computer vision and deep learning, comprising: Obtain visual inspection information of the leather to be inspected in a preset inflated state, wherein the visual inspection information includes light transmission information and elastic deformation information; Information on hidden defects in leather is obtained based on light transmission information, wherein the information on hidden defects in leather texture and information on hidden defects in leather processing are included. The leather texture evaluation value is obtained based on the leather damage information and elastic deformation information. A leather processing evaluation value is generated based on the information on hidden damage and elastic deformation in the leather processing. The overall leather evaluation score is obtained based on the leather texture evaluation value and the leather processing evaluation value. Determine whether the overall evaluation score of the leather exceeds a preset threshold; If the value does not exceed the limit, the leather to be tested is deemed qualified. If the value exceeds the limit, the leather to be tested is deemed unqualified, and leather evaluation information is generated based on the overall leather evaluation score.

[0006] Preferably, the step of obtaining visual inspection information of the leather to be inspected in a preset inflated state includes: The leather to be tested is placed in an air-stretching device, and the leather is subjected to air-stretching treatment based on preset air parameters to form an air-stretched leather state. A surface texture image set is obtained by acquiring surface texture images of the front side of the stretched leather. Transmitted light image sequence was obtained by acquiring transmitted light images of the back side of the stretched leather. Multi-scale texture feature extraction is performed on a set of surface texture images to obtain texture feature representation vectors; Light intensity distribution matrix is ​​obtained by performing light intensity distribution analysis on the transmitted light image sequence; Transmittance information is generated based on texture feature representation vectors and light intensity distribution matrix; Real-time tracking of the edge contour of stretched leather to obtain the edge contour displacement trajectory; Feature point displacement vectors are calculated by matching feature points in a set of surface texture images. Elastic deformation information is generated based on the displacement trajectory of the edge contour and the displacement vector of the feature points.

[0007] Preferably, the step of obtaining information about hidden damage to the leather based on light transmission information includes: The distribution of light intensity attenuation regions is obtained from the light intensity distribution matrix in the light transmission information; The initial suspected dark damage area is obtained based on the distribution of light intensity attenuation area, and the light intensity contrast matrix is ​​obtained based on the initial suspected dark damage area. The contour features of the dark damage area are obtained based on the light intensity contrast matrix; The morphological parameters of the hidden injury region are obtained based on the contour features of the hidden injury region, wherein the morphological parameters of the hidden injury region include the area parameter, perimeter parameter, concavity and convexity parameter and shape complexity parameter of the hidden injury region contour. The morphological parameters of the dark damage area are matched with the preset texture dark damage morphological template and the preset processing dark damage morphological template, and the dark damage area type is determined by the one with the higher matching degree. Obtain information on hidden damage in leather texture and leather processing based on the type of hidden damage area.

[0008] 4. The method for feature recognition of leather products based on the combination of computer vision and deep learning according to claim 3, characterized in that the step of obtaining the distribution of light intensity attenuation region based on the light intensity distribution matrix in the light transmission information includes: Obtain the light intensity value of each grid cell in the light intensity distribution matrix; The light intensity gradient value between adjacent grid cells is obtained based on the light intensity value of each grid cell. The light intensity gradient values ​​are thresholded and filtered, and the set of light intensity gradient values ​​greater than the threshold is generated into a set of abrupt gradient points; By expanding the region centered on the set of abrupt gradient points, a distribution of light intensity attenuation regions is generated.

[0009] Preferably, the step of obtaining the initial suspected dark damage area based on the light intensity attenuation region distribution, and obtaining the light intensity contrast matrix based on the initial suspected dark damage area, includes: Establish a regularly distributed sampling grid within the initial suspected area of ​​hidden injury, and obtain multiple sampling points based on the sampling grid; With each sampling point as the center, multiple ring-shaped regions are constructed outwards to obtain the average light intensity value within each ring-shaped region; The degree of difference between the light intensity value of the central sampling point and the average light intensity value of each layer of the annular region is obtained, and the radial contrast sequence of the sampling point is obtained based on the degree of difference. The radial contrast sequence of each sampling point is normalized to obtain a normalized contrast vector. Based on the spatial location of each sampling point within the initial suspected dark injury area, the corresponding normalized contrast vectors are arranged into a two-dimensional matrix to form the light intensity contrast matrix.

[0010] Preferably, the step of obtaining leather texture hidden damage information and leather processing hidden damage information according to the type of hidden damage area includes: Based on the type of the hidden injury area, the discrimination result information is obtained; When the determination result information is texture dark damage, the position coordinates of the dark damage area are obtained. Based on the position coordinates, the area parameter and the concavity and convexity parameter are extracted from the shape parameters of the dark damage area. The position coordinates, area parameter and concavity and convexity parameter are integrated to generate the leather texture dark damage information. When the judgment result information is a hidden damage in the leather processing, the position coordinates of the hidden damage area are obtained. Based on the position coordinates, the shape complexity parameter is extracted from the shape parameter of the hidden damage area. Based on the position coordinates, the light intensity attenuation amplitude is extracted from the light intensity contrast matrix. The position coordinates, shape complexity parameter and light intensity attenuation amplitude are integrated to generate the hidden damage information of the leather processing.

[0011] Preferably, the step of obtaining the leather texture evaluation value based on the leather scratch information and elastic deformation information includes: The number of hidden defects per unit area is obtained based on the position coordinates in the hidden defect information of the leather texture, and a hidden defect density factor is generated based on the number of hidden defects. The proportion of the dark damage area to the total area of ​​the detection area is obtained based on the area parameter in the dark damage information of the leather texture, and the dark damage area proportion factor is generated based on the proportion of the dark damage area to the total area of ​​the detection area. Based on the elastic deformation information, the displacement vector of the feature point is obtained, and the displacement vector of the feature point is mapped according to its original position on the surface texture image set to generate a displacement field distribution map. The deformation compatibility coefficient is obtained from the displacement field distribution diagram. By comparing the regions with high dark damage density factors with their corresponding texture response intensities, the deformation suppression index of the dark damage regions is obtained. Based on the distribution characteristics of the dark damage and the texture deformation characteristics, the texture-deformation correlation degree is obtained; Based on the texture-deformation correlation, obtain the texture stability index; The leather texture evaluation value is generated by weighted fusion based on the texture stability index and the dark damage distribution characteristics.

[0012] Preferably, the step of generating a leather processing evaluation value based on the leather processing damage information and elastic deformation information includes: The spatial coordinate set of the hidden damage area is obtained based on the position coordinates in the leather processing hidden damage information; Generate a regional deformation response sequence based on the spatial coordinate set of the hidden injury area; The difference in displacement of multiple hidden damage area edge contours before and after stretching is obtained based on the regional deformation response sequence, and contour displacement difference value is generated. The ratio of the maximum value to the minimum value of the contour displacement difference value is obtained based on the distribution of the contour displacement difference value at the edge of the hidden damage area, and deformation gradient factor is generated. The deformation gradient of the hidden damage area is obtained based on the deformation gradient factor. The shape complexity parameter of the hidden damage area is compared with the complexity benchmark value of the preset regular shape to obtain the degree of deviation of the hidden damage area. A shape deviation coefficient is generated based on the degree of deviation. The total shape deviation per unit area is obtained based on the area of ​​the hidden damage area. A structural heterogeneity index is generated based on the total shape deviation. The defect structural heterogeneity is obtained based on the structural heterogeneity index. The density variation degree inside the defect is obtained based on the light intensity attenuation amplitude in the leather processing dark damage information; Based on the deformation gradient of the hidden damage area, the structural heterogeneity of the defect, and the density variability inside the defect, a multi-factor coupling analysis is performed to generate a comprehensive impact index of processing defects. The leather processing evaluation value is generated by mapping and matching the comprehensive impact index of processing defects with the preset processing quality benchmark spectrum.

[0013] Preferably, the step of obtaining the comprehensive leather evaluation score based on the leather texture evaluation value and the leather processing evaluation value includes: The spatial coordinate set of the hidden damage area is obtained based on the position coordinates in the leather processing hidden damage information; Based on the spatial coordinate set of the hidden damage area, the contour deformation data corresponding to multiple hidden damage areas are extracted from the elastic deformation information, and a regional deformation response sequence is generated based on the contour deformation data. The contour displacement of multiple hidden damage areas is obtained according to the regional deformation response sequence, and the deformation intensity parameter is obtained according to the maximum value of the contour displacement. The deformation response degree of multiple hidden damage areas is obtained according to the ratio of the deformation intensity parameter to the preset deformation benchmark. The deviation of the shape of the hidden injury region from the regular geometric shape is obtained based on the shape complexity parameter of the hidden injury region and the preset standard circular complexity parameter. Based on the relationship between the deviation degree and the preset threshold, the structural distortion index of multiple hidden injury regions is obtained. The light intensity attenuation level is divided based on the light intensity attenuation amplitude within the dark damage area. The area ratio of different attenuation level areas within the dark damage area is obtained, and the ratio of the boundary length between each attenuation level area to the total boundary length is obtained based on the area ratio. The density non-uniformity parameter is obtained based on the area ratio and the proportion, and the density variation characterization value of multiple dark injury areas is obtained based on the density non-uniformity parameter. The coupling influence factor of processing defects is obtained based on the deformation response degree; Obtain multiple quality grades and their corresponding ranges of coupling influence factors, and obtain the final leather processing evaluation value based on the ranges of coupling influence factors and the coupling influence factors of processing defects.

[0014] Preferably, the step of generating leather assessment information based on the overall leather assessment degree includes: Based on the comprehensive leather evaluation score, the comprehensive leather evaluation score is compared with multiple preset quality grade threshold ranges to obtain the comparison results; The corresponding quality level is obtained based on the comparison results, and a quality grading identifier is obtained based on the quality level. Based on the quality grading identifier, the density factor and the area ratio factor of hidden damage are extracted from the leather texture evaluation value, and the coupling influence factor of processing defects is extracted from the leather processing evaluation value. Based on the aforementioned hidden defect density factor, hidden defect area ratio factor, and processing defect coupling influence factor, a fusion process is performed to generate a comprehensive defect influence degree. The overall impact of the defect is compared with a preset defect severity threshold, and the comparison result is obtained. The defect severity level is then obtained based on the comparison result. A defect severity classification is generated based on the aforementioned defect severity level; Based on the defect severity classification, spatial location information of the hidden damage area, and defect type information, a repair urgency index for multiple hidden damage areas is obtained; and the multiple hidden damage areas are sorted according to the repair urgency index to generate a defect repair priority sequence. Based on the quality grading identifier, the overall impact of defects, the classification of defect severity, and the priority sequence of defect repair, information is integrated to generate structured leather assessment information.

[0015] The beneficial effects of this application are as follows: This invention effectively integrates surface visual features and internal structural information by simultaneously acquiring surface texture images and back-side transmitted light images of leather under inflated and stretched conditions. Transmitted light imaging can penetrate the material and clearly reveal hidden defects that are difficult to detect by traditional methods, such as internal dark scratches, uneven density, and microcracks. This overcomes the detection blind spots caused by reflection and texture interference in single-surface imaging, and significantly improves the coverage and accuracy of defect identification. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a method flow according to an embodiment of this application.

[0017] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] like Figure 1 As shown, this application provides a feature recognition method for leather products based on a combination of computer vision and deep learning, including: S1. Obtain visual detection information of the leather to be tested in a preset inflated state, wherein the visual detection information includes light transmission information and elastic deformation information; S2. Obtain leather hidden damage information based on light transmission information, wherein the leather hidden damage information includes leather texture hidden damage information and leather processing hidden damage information; S3. Obtain leather texture evaluation value based on the leather damage information and elastic deformation information; S4. Generate a leather processing evaluation value based on the leather processing damage information and elastic deformation information; S5. Obtain the comprehensive leather evaluation score based on the leather texture evaluation value and the leather processing evaluation value; S6. Determine whether the overall evaluation score of the leather exceeds a preset threshold; If the value does not exceed the limit, the leather to be tested is deemed qualified. If the value exceeds the limit, the leather to be tested is deemed unqualified, and leather evaluation information is generated based on the overall leather evaluation score.

[0020] As described in steps S1-S6 above, this invention acquires visual inspection information of leather under a preset inflated state, particularly including light transmission information and elastic deformation information. Light transmission information effectively reveals hidden defects such as internal flaws and uneven density in the leather, compensating for the inability of surface optical inspection to penetrate the material's interior. Elastic deformation information quantifies the dynamic response of the leather under stress, allowing defects that only manifest or worsen under stress (such as microcrack propagation and weak fiber areas) to be captured. The acquired flaw information is finely categorized into leather texture flaw information and leather processing flaw information. This categorization, based on the different causes and morphological characteristics of defects, enables the system to distinguish between defects originating from the leather's natural texture structure (such as scars and blood vessel marks) and defects introduced during processing (such as uneven tanning and mechanical damage). This categorization not only helps trace the root cause of quality problems but also independently generates leather texture evaluation values ​​and leather processing evaluation values ​​based on the categorized flaw information and elastic deformation information. Step S3 analyzes the distribution density and area ratio of texture defects, and combines this with the deformation response under tension (such as the consistency of the displacement field and deformation inhibition effect) to comprehensively evaluate the integrity, uniformity, and structural stability of the leather's natural texture. Step S4 quantifies the deformation gradient, structural heterogeneity, and internal density variation of the processed defect area to comprehensively assess the impact of processing technology on the leather's mechanical properties and internal structure. The evaluation values ​​from these two dimensions are merged to obtain the overall leather evaluation score, comprehensively considering the texture stability, the severity of processing defects, and their spatial distribution influence, ultimately outputting a comprehensive, quantitative, and unified overall quality index. The overall leather evaluation score is judged based on a preset threshold, and structured leather evaluation information is generated. The complex multi-dimensional analysis results are transformed into a clear "qualified / unqualified" judgment. For unqualified products, the system not only provides a comprehensive quality level but also generates a detailed evaluation report including defect severity classification, specific spatial location, defect type, and a priority sequence based on the repair urgency index. In one embodiment, the step of obtaining visual detection information of the leather to be detected in a preset inflated state includes: S101. Place the leather to be tested in the air-stretching device and perform air-stretching treatment on the leather based on preset air parameters to form leather in an air-stretched state. S102. Collect surface texture images of the front side of the stretched leather to obtain a surface texture image set; S103. Acquire transmitted light images of the back side of the stretched leather to obtain a sequence of transmitted light images. S104. Perform multi-scale texture feature extraction on the surface texture image set to obtain texture feature representation vector; S105. Perform light intensity distribution analysis on the transmitted light image sequence to obtain the light intensity distribution matrix; S106. Generate light transmission information based on texture feature representation vector and light intensity distribution matrix; S107. Real-time tracking of the edge contour of the stretched leather to obtain the edge contour displacement trajectory. S108. Perform feature point matching on the surface texture image set and calculate the feature point displacement vector; S109. Generate elastic deformation information based on the displacement trajectory of the edge contour and the displacement vector of the feature point.

[0021] As described in steps S101-S109 above, the leather to be inspected is placed in an air-stretching device. Based on preset air-stretching parameters, the leather is subjected to air-stretching treatment to form an air-stretched leather state. This ensures that the leather is inspected under controlled and uniform stretching, simulating the stress conditions in actual use and making potential defects more easily visible under deformation. The air-stretching device may include a blower, which fixes the leather to be inspected above it and applies a uniform vertical or radial stretching force to the leather by controlling the air pressure. Preset air-stretching parameters may include air pressure, air-stretching rate, and holding time. Surface texture images are acquired from the front of the stretched leather to obtain a surface texture image set. This set is used to acquire visual information about the leather surface, including texture, color, and gloss, for subsequent texture feature extraction and defect identification. A high-resolution industrial camera can be used with a ring light source or diffused light source to acquire images from directly above the leather to obtain clear, uniformly illuminated surface texture images. Transmitted light image acquisition is performed on the back of stretched leather to obtain a sequence of transmitted light images. This technique allows for the acquisition of visual information about the leather's internal structure and density variations, aiding in the detection of internal defects such as dark spots and density inconsistencies. A uniform surface light source can be placed below the back of the leather, with an industrial camera positioned above it to capture images and record the intensity distribution of light passing through the leather, forming a sequence of transmitted light images. Multi-scale texture feature extraction is then performed on the surface texture image set to obtain texture feature representation vectors. Traditional image processing methods such as Gabor filter banks, wavelet transforms, or local binary modes can be used to filter and calculate features at multiple scales and directions to generate texture feature representation vectors. Light intensity distribution analysis is then performed on the transmitted light image sequence to obtain a light intensity distribution matrix. This aims to quantify the intensity distribution of light in the transmitted light image sequence, reflecting the uniformity and density variations within the leather, providing a data foundation for subsequent acquisition of dark spot information. Each frame of a transmitted light image sequence can be converted to grayscale, and the grayscale value of each pixel can be calculated and organized into a two-dimensional matrix to form a light intensity distribution matrix. This matrix integrates leather surface texture features and internal light intensity distribution information to create more comprehensive and multi-dimensional light transmission information, more accurately reflecting the overall light transmission characteristics of the leather. Texture feature vectors and the light intensity distribution matrix can be concatenated, directly combining them into a longer vector or a higher-dimensional matrix as light transmission information. Real-time tracking of the edge contour of stretched leather obtains its displacement trajectory, which monitors changes in the overall shape and size of the leather during stretching, providing data for assessing the macroscopic elastic deformation of the leather. An image processing-based edge detection algorithm combined with a contour tracking algorithm can be used to extract and track the edge contour of the leather in real time from a continuously acquired image sequence, recording its displacement trajectory over time.Feature point matching and displacement vector calculation of surface texture images aim to identify local deformation on the leather surface. By tracking the positional changes of specific feature points before and after stretching, microscopic elastic deformation is quantified. Classic feature point detection and description algorithms such as SIFT, SURF, and ORB can be used to extract and match feature points in the surface texture images before and after stretching, and then calculate the displacement between the matched feature point pairs to generate feature point displacement vectors. Elastic deformation information is generated based on edge contour displacement trajectories and feature point displacement vectors. This aims to integrate macroscopic edge contour deformation and microscopic feature point deformation to form comprehensive elastic deformation information for accurately assessing the deformation characteristics of leather under stress. Edge contour displacement trajectories and feature point displacement vectors can be fused to generate elastic deformation information describing the overall elastic deformation of the leather.

[0022] In one embodiment, the step of obtaining information about leather blemishes based on light transmission information includes: S201. Obtain the distribution of light intensity attenuation region based on the light intensity distribution matrix in the light transmission information; S202. Obtain the initial suspected dark damage area based on the distribution of light intensity attenuation area, and obtain the light intensity contrast matrix based on the initial suspected dark damage area. S203. The light intensity contrast matrix is ​​binarized to generate a mask for the dark damage area. Edge detection is performed on the mask to obtain the edge contour of the dark damage area. Polygon approximation is performed on the edge contour of the dark damage area to obtain the contour features of the dark damage area. S204. Obtain the shape parameters of the hidden injury area based on the contour features of the hidden injury area, wherein the shape parameters of the hidden injury area include the area parameter, perimeter parameter, concavity and convexity parameter and shape complexity parameter of the hidden injury area contour. S205. Match the morphological parameters of the dark damage area with the preset texture dark damage morphological template and the preset processing dark damage morphological template, and determine the dark damage area type based on the higher matching degree. S206. Obtain information on leather texture defects and leather processing defects based on the type of defect area.

[0023] As described in steps S201-S206 above, this invention identifies regions of light attenuation by analyzing the light intensity distribution matrix in the light transmission information. These regions are preliminary indicators of potential hidden defects. This step utilizes the influence of the hidden defect region on light transmittance, effectively separating abnormal regions from the background. Next, based on these initially identified light intensity attenuation regions, initial suspected hidden defect regions are further obtained, and the salience of the hidden defect features is enhanced by calculating the light intensity contrast of these regions. The generation of the light intensity contrast matrix makes the boundaries and internal structure of the hidden defect regions clearer. The light intensity contrast matrix is ​​binarized to generate a hidden defect region mask, completely separating the hidden defect regions from the background. Based on this, edge detection is performed to obtain the edge contours of the hidden defect regions, and the contours are simplified using polygon approximation to extract their key geometric features. These contour features are the basis for quantifying the morphology of the hidden defects. Based on these contour features, the morphological parameters of the hidden defect regions are calculated, including area parameters, perimeter parameters, concavity / convexity parameters, and shape complexity parameters. These parameters objectively describe the physical shape and structure of the hidden defects, providing a quantitative basis for distinguishing different types of hidden defects. These quantified morphological parameters are matched against preset templates for textured and processed leather defects. For example, textured defects may appear as irregular shapes related to the natural grain of the leather, while processed defects may appear as more regular scratches or indentations caused by external forces. By comparing the matching degree, the system can intelligently determine the type of defect. Finally, based on the determined defect type, the relevant morphological parameters, location information, and other features are structurally integrated to generate leather textured defect information and leather processed defect information, respectively. This categorized and structured information output enables subsequent leather quality assessments to be more precise and targeted at different types of defects, thereby significantly improving the accuracy of the entire assessment.

[0024] In one embodiment, the step of obtaining the light intensity attenuation region distribution based on the light intensity distribution matrix in the light transmission information includes: S2011. Obtain the light intensity value of each grid cell in the light intensity distribution matrix; S2012. Obtain the light intensity gradient value between adjacent grid cells based on the light intensity value of each grid cell; S2013. Threshold screening is performed on the light intensity gradient values, and the set of light intensity gradient values ​​greater than the threshold is used to generate a set of abrupt gradient points. S2014. Expand the region centered on the abrupt gradient point set to generate the light intensity attenuation region distribution.

[0025] As described in steps S2011-S2014 above, this invention obtains the light intensity value of each grid cell in the light intensity distribution matrix. By calculating the light intensity gradient values ​​between adjacent grid cells, the drastic change in light intensity in space can be quantified, thereby effectively capturing abrupt changes in translucency caused by internal structure or defects in the leather. To eliminate noise interference and insignificant minor changes, these light intensity gradient values ​​are thresholded, retaining only those significant gradient points that indicate potential hidden defects, forming a set of abrupt gradient points. Finally, a region expansion operation is performed centered on these abrupt gradient points, connecting the discrete abrupt points into continuous regions, thus forming a complete distribution of light intensity attenuation regions. By introducing light intensity gradient analysis and thresholding, noise interference and insignificant light intensity fluctuations are effectively filtered out, making the identified abrupt gradient points more representative.

[0026] In one embodiment, the step of obtaining an initial suspected dark damage area based on the light intensity attenuation region distribution, and obtaining a light intensity contrast matrix based on the initial suspected dark damage area, includes: S2021. Establish a regularly distributed sampling grid within the initial suspected area of ​​hidden injury, and obtain multiple sampling points based on the sampling grid; S2022. Construct multiple ring-shaped regions outward from each sampling point as the center, and obtain the average light intensity value within each ring-shaped region. S2023. Obtain the degree of difference between the light intensity value of the central sampling point and the average light intensity value of each layer of the annular region, and obtain the radial contrast sequence of the sampling point based on the degree of difference. S2024. Normalize the radial contrast sequence of each sampling point to obtain a normalized contrast vector. S2025. Based on the spatial position of each sampling point within the initial suspected dark injury area, arrange the corresponding normalized contrast vectors into a two-dimensional matrix to form the light intensity contrast matrix.

[0027] As described in steps S2021-S2025 above, this invention establishes a regularly distributed sampling grid within the initial suspected dark flaw area and acquires multiple sampling points based on this grid. This step ensures comprehensive coverage and uniform sampling of the potential dark flaw area, avoiding deviations caused by local illumination variations or random sampling, and laying a data foundation for subsequent detailed analysis. Subsequently, multiple ring-shaped regions are constructed outward from each sampling point, and the average light intensity value within each ring-shaped region is acquired. This multi-ring structure design allows the system to capture the attenuation or change pattern of light intensity from the center to the periphery at different spatial scales, thereby more sensitively identifying subtle features of the dark flaw boundary, rather than relying solely on local contrast at a single scale. Next, by acquiring the degree of difference between the light intensity value of the central sampling point and the average light intensity value of each ring-shaped region, a radial contrast sequence for that sampling point is generated. This sequence quantifies the radial gradient of light intensity variation, effectively distinguishing real dark flaw areas from background noise or illumination artifacts. To eliminate the dimensional differences in the absolute values ​​of light intensity between different sampling points and ensure the comparability of contrast data, the radial contrast sequence of each sampling point is normalized to obtain a normalized contrast vector. Finally, based on the spatial position of each sampling point within the initial suspected dark spot area, the corresponding normalized contrast vectors are arranged into a two-dimensional matrix to form a light intensity contrast matrix.

[0028] In one embodiment, the step of obtaining leather texture hidden damage information and leather processing hidden damage information according to the type of hidden damage area includes: S2061. Obtain the discrimination result information based on the type of the hidden damage area; When the determination result information is texture dark damage, the position coordinates of the dark damage area are obtained. Based on the position coordinates, the area parameter and the concavity and convexity parameter are extracted from the shape parameters of the dark damage area. The position coordinates, area parameter and concavity and convexity parameter are integrated to generate the leather texture dark damage information. When the judgment result information is a hidden damage in the leather processing, the position coordinates of the hidden damage area are obtained. Based on the position coordinates, the shape complexity parameter is extracted from the shape parameter of the hidden damage area. Based on the position coordinates, the light intensity attenuation amplitude is extracted from the light intensity contrast matrix. The position coordinates, shape complexity parameter and light intensity attenuation amplitude are integrated to generate the hidden damage information of the leather processing.

[0029] As described in step S2061 above, this invention solves the problems of inaccurate and inefficient information extraction in traditional methods by classifying the types of hidden damage areas and extracting different types of hidden damage information in a targeted manner based on the classification results. Specifically, after obtaining leather hidden damage information (including leather texture hidden damage information and leather processing hidden damage information), the detected hidden damage areas are first classified to determine whether they are texture hidden damage or processing hidden damage. This classification result is the basis for subsequent information extraction. When the classification result indicates a texture hidden damage, the focus is on extracting parameters related to the natural texture defects of the leather, that is, obtaining the position coordinates of the hidden damage area, and accurately extracting the area parameters and concavity / convexity parameters from the previously obtained morphological parameters of the hidden damage area. These parameters are integrated to form leather texture hidden damage information, which focuses on describing the size and surface irregularity of the texture defects. Conversely, when the classification result indicates a processing hidden damage, the focus shifts to extracting parameters related to the processing defects. Similarly, the position coordinates of the hidden damage area are obtained, but the shape complexity parameter is extracted from the morphological parameters of the hidden damage area, and the light intensity attenuation amplitude is extracted from the light intensity contrast matrix. These parameters are integrated to form information on hidden defects in leather processing, which focuses more on the geometric anomalies and internal structural changes of the defects. This type-driven parameter extraction mechanism ensures the relevance and effectiveness of the acquired information, avoids interference from irrelevant data, and makes subsequent leather evaluation more accurate and efficient.

[0030] In one embodiment, the step of obtaining the leather texture evaluation value based on the leather scratch information and elastic deformation information includes: S301. Obtain the number of hidden defects per unit area based on the position coordinates in the hidden defect information of the leather texture, and generate a hidden defect density factor based on the number of hidden defects. S302. Obtain the proportion of the dark damage area to the total area of ​​the detection area based on the area parameter in the dark damage information of the leather texture, and generate the dark damage area proportion factor according to the proportion of the dark damage area to the total area of ​​the detection area. S303. Obtain the feature point displacement vector according to the elastic deformation information, and map the feature point displacement vector according to its original position on the surface texture image set to generate a displacement field distribution map. S304. Obtain the deformation compatibility coefficient based on the displacement field distribution diagram; S305. Compare the regions with high dark damage density factors with the corresponding texture response intensities to obtain the deformation suppression index of the dark damage regions. S306. Obtain the texture-deformation correlation degree based on the dark damage distribution characteristics and the texture deformation characteristics; S307. Obtain the texture stability index based on the texture-deformation correlation degree; S308. The leather texture evaluation value is generated by weighted fusion based on the texture stability index and the dark damage distribution characteristics.

[0031] As described in steps S301-S308 above, this invention quantifies the spatial density of surface texture defects on leather. The location coordinates can be the coordinates of the center point or bounding box of the defect area identified by an image processing algorithm. The defect area ratio factor is used to evaluate the coverage and scale of texture defects on the overall leather area. The area parameter refers to the actual physical area or pixel area of ​​each identified defect area. The total detection area is the overall effective area of ​​the leather to be detected. The defect area ratio factor is the ratio of the total area of ​​all texture defects to the total area of ​​the detection area, usually expressed as a percentage, reflecting the macroscopic impact of defects and visualizing the deformation of the leather under tension. Elastic deformation information includes the deformation data of the leather after being subjected to force, such as the positional changes of feature points obtained through image registration or optical flow before and after tension. The feature point displacement vector represents the direction and magnitude of displacement of each feature point from its original position to its deformed position. The displacement field distribution map arranges and visualizes these displacement vectors according to their spatial positions on the original image, forming a two-dimensional vector field that visually displays the deformation trend of various points on the leather surface. Texture response intensity is used to evaluate the uniformity and response characteristics of local areas of the leather under deformation. Region division involves dividing the displacement field distribution map into several sub-regions. These sub-regions can be regular grids or adaptive regions based on image content (such as texture features). Orientation consistency refers to the degree to which the directions of all displacement vectors deviate from the average direction within each sub-region. Texture response intensity is an index calculated based on orientation consistency; higher orientation consistency generally indicates more uniform texture deformation and higher response intensity in that region. The deformation coordination coefficient is a quantitative indicator of the difference in texture response intensity between adjacent regions. The smaller the difference, the more coordinated the deformation response of adjacent regions, and the more uniform the overall deformation behavior of the leather. The deformation suppression index is a quantitative indicator representing the degree of decrease in texture response intensity in areas with high density of hidden damage compared to areas with no or low density of hidden damage. A higher inhibition index indicates that minor scratches have a significant negative impact on the elastic deformation of leather. The texture stability index is a comprehensive assessment of the leather texture's ability to maintain its structural integrity and functional properties under stress and deformation. A higher correlation generally means a more significant impact of minor scratches on deformation, or a stronger ability of deformation to reveal minor scratches, thus allowing for a more accurate assessment of the texture's intrinsic stability. The leather texture evaluation value is generated by weighted fusion of the texture stability index and minor scratch distribution characteristics. This step is crucial for generating the final leather texture evaluation value, as it integrates the intrinsic stability of the texture and the macroscopic distribution influence of minor scratches. Weighted fusion refers to linearly or non-linearly combining the texture stability index and minor scratch distribution characteristics (such as minor scratch density factor and minor scratch area ratio factor) according to preset weights to obtain a single comprehensive evaluation value. In one embodiment, the step of generating a leather processing evaluation value based on the leather processing damage information and elastic deformation information includes: S401. Obtain the spatial coordinate set of the hidden damage area based on the position coordinates in the leather processing hidden damage information; S402. Based on the spatial coordinate set of the hidden damage area, obtain local contour deformation data corresponding to multiple hidden damage areas from the edge contour displacement trajectory in the elastic deformation information, and generate a regional deformation response sequence based on the local contour deformation data corresponding to multiple hidden damage areas. S403. Based on the regional deformation response sequence, obtain the difference in displacement of multiple hidden damage area edge contours before and after stretching, generate contour displacement difference value, obtain the ratio of the maximum value to the minimum value based on the distribution of the contour displacement difference value at the edge of the hidden damage area, generate deformation gradient factor, and obtain the deformation gradient of the hidden damage area based on the deformation gradient factor. S404. Compare the shape complexity parameter of the hidden damage area with the complexity benchmark value of the preset regular shape to obtain the deviation degree of the hidden damage area, generate a shape deviation coefficient based on the deviation degree, obtain the total shape deviation per unit area based on the area of ​​the hidden damage area, generate a structural heterogeneity index based on the total shape deviation, and obtain the defect structural heterogeneity based on the structural heterogeneity index. S405. Based on the light intensity attenuation amplitude in the leather processing dark damage information, obtain the density variation degree inside the defect; S406. Based on the deformation gradient of the hidden damage area, the structural heterogeneity of the defect, and the internal density variability of the defect, perform multi-factor coupling analysis to generate a comprehensive impact index of processing defects. S407. Based on the comprehensive impact index of processing defects and the preset processing quality benchmark spectrum, a leather processing evaluation value is generated by mapping and matching.

[0032] As described in steps S401-S407 above, the spatial coordinate set of the hidden damage area in this invention represents the geometric location information of these defects on the leather surface, typically expressed in pixel coordinates, physical coordinates, or region bounding boxes. This provides a precise analytical range and data index for subsequent deformation analysis, structural heterogeneity assessment, and density variability calculation. Local contour deformation data refers to the displacement or shape change information of the edge of the hidden damage area before and after stretching, while the regional deformation response sequence is the result of organizing and expressing these deformation data in a certain order or logic. The contour displacement difference value is the difference in displacement between different points on the edge of the hidden damage area before and after stretching, reflecting the relative change in local deformation. The deformation gradient factor is a dimensionless parameter calculated based on these difference values, used to characterize the severity of deformation changes. The deformation gradient of the hidden damage area is the result of further processing or quantification of the deformation gradient factor, used to comprehensively describe the deformation non-uniformity of the hidden damage area. The shape complexity parameter is an index describing the geometric shape complexity of the hidden damage area. The shape deviation coefficient is a quantified result after comparing the defect shape complexity with a preset rule shape complexity benchmark value, reflecting the irregularity of the shape. Structural heterogeneity index combines shape deviation coefficient and area of ​​hidden damage region to measure the total amount of structural irregularities per unit area. Defect structural heterogeneity is a further abstraction and quantification of structural heterogeneity index, used to comprehensively assess the degree of structural anomaly of defects. Light intensity attenuation amplitude is an indicator reflecting the degree of intensity attenuation when light penetrates leather in leather processing hidden damage information, usually related to the material's thickness, density, or internal structural defects. Defect internal density variability is a quantitative index calculated based on light intensity attenuation amplitude, used to describe the non-uniformity of density distribution within the hidden damage region. It reveals the physical structural anomalies of the material within the hidden damage region, such as internal voids, loose fibers, or uneven density. Multi-factor coupling analysis is a method that integrates and analyzes multiple interrelated or interacting factors. The processing defect comprehensive impact index is the result of this analysis, representing the overall severity and potential impact of defects. It provides a unified and comprehensive defect assessment standard, avoiding the one-sidedness of a single indicator; the preset processing quality benchmark spectrum is a series of reference standards defining different quality levels, usually including the range or threshold of the processing defect comprehensive impact index corresponding to each quality level. Mapping matching refers to comparing the calculated index with a benchmark spectrum to determine its quality level. It fully utilizes information on hidden defects and elastic deformation in leather processing to perform multi-dimensional and refined quantitative analysis of processing defects in leather products. Specifically, by accurately obtaining the spatial coordinates of the hidden defect area and combining this with elastic deformation information to capture its local deformation response under tension, the mechanical behavior characteristics of the hidden defect area can be revealed. Furthermore, by calculating the deformation gradient of the hidden defect area, the non-uniformity of deformation at the defect edge can be quantified, thereby reflecting the local discontinuities in the material's mechanical properties.By assessing the structural heterogeneity of areas with hidden defects, the geometric irregularities of processing defects can be identified and quantified, helping to distinguish different types of processing defects. Obtaining the density variability within defects reveals physical structural anomalies in the material within these areas. These multi-dimensional features (deformation gradient, structural heterogeneity, and density variability) are organically integrated into a multi-factor coupling analysis to generate a comprehensive processing defect impact index. This comprehensive assessment method avoids the limitations of single indicators and more comprehensively and accurately reflects the overall impact of processing defects on the overall performance of the leather. Finally, by mapping and matching with a pre-defined processing quality benchmark spectrum, objective and quantitative leather processing evaluation values ​​can be generated.

[0033] In one embodiment, the step of obtaining the overall leather assessment score based on the leather texture assessment value and the leather processing assessment value includes: S501. Obtain the spatial coordinate set of the hidden damage area based on the position coordinates in the leather processing hidden damage information; S502. Based on the spatial coordinate set of the hidden damage area, extract the contour deformation data corresponding to multiple hidden damage areas from the elastic deformation information, and generate a regional deformation response sequence based on the contour deformation data. S503. Obtain the contour displacement of multiple hidden damage areas according to the regional deformation response sequence, obtain the deformation intensity parameter according to the maximum value of the contour displacement, and obtain the deformation response degree of multiple hidden damage areas according to the ratio of the deformation intensity parameter to the preset deformation benchmark. S504. Obtain the deviation between the shape of the hidden injury area and the regular geometric shape based on the shape complexity parameter of the hidden injury area and the preset standard circular complexity parameter, and obtain the structural distortion index of multiple hidden injury areas based on the relationship between the deviation degree and the preset threshold. S505. Divide the light intensity attenuation level based on the light intensity attenuation amplitude inside the dark damage area, obtain the area ratio of different attenuation level areas in the dark damage area, and obtain the ratio of the boundary length between each attenuation level area to the total boundary length based on the area ratio. S506. Obtain density non-uniformity parameters based on the area ratio and the proportion, and obtain density variation characterization values ​​for multiple dark injury areas based on the density non-uniformity parameters. S507. Normalize the deformation response to obtain normalized deformation intensity, normalize the structural distortion index to obtain normalized structural distortion degree, normalize the density variation characterization value to obtain normalized density variation degree, and perform weighted fusion of the normalized deformation intensity, normalized structural distortion degree, and normalized density variation degree according to preset coupling weights to obtain processing defect coupling influence factor. S508. Obtain multiple quality grades and their corresponding coupling influence factor ranges, match the processing defect coupling influence factor with the coupling influence factor ranges corresponding to each quality grade, determine the preliminary quality grade corresponding to the leather processing evaluation value based on the matching results, fine-tune the preliminary quality grade based on the degree to which the processing defect coupling influence factor deviates from the current grade center value, and generate the final leather processing evaluation value.

[0034] As described in steps S501-S508 above, the spatial coordinate set of the hidden damage area refers to the set of location information of the identified processing hidden damage areas on the leather product in two-dimensional or three-dimensional space. It accurately locates the defect, providing a spatial reference for subsequent deformation, structure, and density analysis. The contour deformation data describes the boundary shape of the hidden damage area before and after stretching. It captures the dynamic response of the defect under stress. This data may include: a set of displacement vectors of key points on the contour of the hidden damage area; a regional deformation response sequence recording an ordered set of contour deformation data of the hidden damage area changing over time or with varying degrees of stretching; revealing the dynamic process of deformation behavior in the hidden damage area; and contour displacement quantifies the magnitude of displacement of each point or the entire contour of the hidden damage area during stretching; measuring the deformation amplitude of the hidden damage area. The deformation intensity parameter reflects the intensity of contour deformation in the hidden damage area, usually based on the maximum value of the contour displacement; highlighting the most severe parts or degrees of deformation in the hidden damage area; and a preset deformation benchmark used to compare and evaluate the standard values ​​or ranges of the deformation intensity parameter; providing an objective reference to determine whether the deformation of the hidden damage area is abnormal. Deformation response is a quantitative index obtained by comparing the deformation intensity parameter of the hidden damage area with a preset deformation benchmark. It standardizes the deformation intensity to make it comparable. The shape complexity parameter describes the geometric complexity of the hidden damage area. It quantifies the irregularity of the defect shape. The preset standard circular complexity parameter is the complexity value of an ideal circular shape used for comparison with the shape complexity parameter of the hidden damage area. It provides an ideal geometric reference. Deviation is the degree of difference between the shape complexity parameter of the hidden damage area and the preset standard circular complexity parameter. It quantifies the degree of deviation of the defect shape from the ideal shape. A preset threshold is used to determine whether the deviation reaches the critical value of the structural distortion standard. It distinguishes between normal shape fluctuations and structural distortion. The structural distortion index is an index that quantifies the degree of structural distortion in the hidden damage area based on the relationship between the deviation and the preset threshold. It assesses the degree of abnormality of the defect shape. The light intensity attenuation amplitude describes the degree of reduction in light transmittance within the hidden damage area. It reflects the density or internal structural anomaly of the hidden damage area. The light intensity attenuation level divides the interior of the hidden damage area into different light transmittance categories according to the light intensity attenuation amplitude. The analysis of density inhomogeneity within defects is refined. Area proportion represents the ratio of regions with different light intensity attenuation levels within the entire dark defect area. The distribution of different density regions within the defect is analyzed. The ratio of boundary length to total boundary length is the ratio of the boundary length between regions with different light intensity attenuation levels within the dark defect area to the total external boundary length of the dark defect area. The fractal characteristics or complexity of the internal structure of the defect are quantified. The density inhomogeneity parameter, combining area proportion and boundary length ratio, characterizes the uniformity or heterogeneity of density distribution within the dark defect area. The complexity of the internal structure of the defect is quantified. The density variation characterization value is a comprehensive index of density variation within the dark defect area, extracted based on the density inhomogeneity parameter. A unified density variation assessment is provided.

[0035] This invention, by introducing precise acquisition of the spatial coordinate set of the hidden damage area and extracting contour deformation data based on elastic deformation information to generate a regional deformation response sequence, enables the dynamic and quantitative capture of the physical behavior of defects under stress, thereby more accurately assessing their deformation intensity. Simultaneously, by quantifying the deviation of the defect shape complexity from the standard geometry and combining information such as light intensity attenuation level, area ratio, and boundary proportion, the structural distortion and internal density inhomogeneity of the defect are comprehensively revealed. These multi-dimensional and refined feature analyses, combined with normalization processing and weighted fusion mechanisms, generate a more comprehensive and reliable coupling influence factor for processing defects. Based on this, through matching with a preset quality grade range and fine-tuning based on the degree of deviation, a highly accurate leather processing evaluation value is finally generated.

[0036] In one embodiment, the step of generating leather assessment information based on the overall leather assessment degree includes: S601. Based on the comprehensive leather evaluation score, compare the comprehensive leather evaluation score with multiple preset quality grade threshold ranges to obtain the comparison results; S602. Obtain the corresponding quality level based on the comparison result, and obtain the quality grading identifier based on the quality level; S603. Based on the quality grading identifier, extract the dark damage density factor and dark damage area ratio factor from the leather texture evaluation value, and extract the processing defect coupling influence factor from the leather processing evaluation value. S604. Based on the aforementioned hidden defect density factor, hidden defect area ratio factor, and processing defect coupling influence factor, perform fusion processing to generate a comprehensive defect influence degree. S605. Compare the overall impact of the defect with a preset defect severity threshold and obtain the comparison result, and obtain the defect severity level based on the comparison result; S606. Generate a defect severity classification based on the aforementioned defect severity level; S607. Based on the defect severity classification, the spatial location information of the hidden damage area, and the defect type information, obtain the repair urgency index of multiple hidden damage areas; and sort the multiple hidden damage areas according to the repair urgency index to generate a defect repair priority sequence. S608. Based on the quality grading identifier, comprehensive impact of defects, classification of defect severity, and priority sequence of defect repair, information is integrated to generate structured leather evaluation information.

[0037] As described in steps S601-S608 above, this invention performs in-depth analysis on the comprehensive leather assessment obtained in the preceding methods to generate highly instructive structured leather assessment information. First, after receiving the comprehensive leather assessment, the system precisely compares it with multiple preset quality grade threshold ranges. The comparison results are then used to obtain specific quality grades and corresponding quality grading identifiers. Based on the determined quality grading identifiers, key factors such as the density factor of hidden damage, the proportion factor of hidden damage area, and the coupling influence factor of processing defects are selectively extracted from the leather texture assessment value and leather processing assessment value. This selective extraction mechanism avoids interference from irrelevant information. These extracted factors are then fused to generate a unified comprehensive defect influence degree, which comprehensively quantifies the overall defect severity of the leather. The comprehensive defect influence degree is compared with preset defect severity thresholds to determine the severity level of the defect. This level is then transformed into a more descriptive defect severity classification, making the nature of the defect more intuitively understandable. It goes beyond simply classifying defects; it combines defect severity classification, spatial location information of hidden damage areas, and defect type information to calculate a repair urgency index for multiple hidden damage areas. This index comprehensively considers the severity, location sensitivity, and type impact of the defect. Finally, all hidden damage areas are ranked according to their repair urgency index, generating a defect repair priority sequence to ensure optimal allocation of repair resources and maximized efficiency. Finally, this solution systematically integrates all the above key information, including quality grading identifiers, comprehensive defect impact, defect severity classification, and defect repair priority sequence, to generate structured leather assessment information.

[0038] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0039] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0040] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent results or equivalent process transformations made based on the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for feature recognition of leather products based on a combination of computer vision and deep learning, characterized in that, include: Obtain visual inspection information of the leather to be inspected in a preset inflated state, wherein the visual inspection information includes light transmission information and elastic deformation information; Information on hidden defects in leather is obtained based on light transmission information, wherein the information on hidden defects in leather texture and information on hidden defects in leather processing are included. The leather texture evaluation value is obtained based on the leather damage information and elastic deformation information. A leather processing evaluation value is generated based on the information on hidden damage and elastic deformation in the leather processing. The overall leather evaluation score is obtained based on the leather texture evaluation value and the leather processing evaluation value. Determine whether the overall evaluation score of the leather exceeds a preset threshold; If the value does not exceed the limit, the leather to be tested is deemed qualified. If the value exceeds the limit, the leather to be tested is deemed unqualified, and leather evaluation information is generated based on the overall leather evaluation score.

2. The method for feature recognition of leather products based on a combination of computer vision and deep learning according to claim 1, characterized in that, The step of obtaining visual inspection information of the leather to be inspected in a preset inflated state includes: The leather to be tested is placed in an air-stretching device, and the leather is subjected to air-stretching treatment based on preset air parameters to form an air-stretched leather state. A surface texture image set is obtained by acquiring surface texture images of the front side of the stretched leather. Transmitted light image sequence was obtained by acquiring transmitted light images of the back side of the stretched leather. Multi-scale texture feature extraction is performed on a set of surface texture images to obtain texture feature representation vectors; Light intensity distribution matrix is ​​obtained by performing light intensity distribution analysis on the transmitted light image sequence; Transmittance information is generated based on texture feature representation vectors and light intensity distribution matrix; Real-time tracking of the edge contour of stretched leather to obtain the edge contour displacement trajectory; Feature point displacement vectors are calculated by matching feature points in a set of surface texture images. Elastic deformation information is generated based on the displacement trajectory of the edge contour and the displacement vector of the feature points.

3. The method for feature recognition of leather products based on a combination of computer vision and deep learning according to claim 1, characterized in that, The step of obtaining information about hidden damage to leather based on light transmission information includes: The distribution of light intensity attenuation regions is obtained from the light intensity distribution matrix in the light transmission information; The initial suspected dark damage area is obtained based on the distribution of light intensity attenuation area, and the light intensity contrast matrix is ​​obtained based on the initial suspected dark damage area. The contour features of the dark damage area are obtained based on the light intensity contrast matrix; The morphological parameters of the hidden injury region are obtained based on the contour features of the hidden injury region. The morphological parameters of the hidden injury region include the area parameter, perimeter parameter, concavity and convexity parameter, and shape complexity parameter of the hidden injury region contour. The morphological parameters of the dark damage area are matched with the preset texture dark damage morphological template and the preset processing dark damage morphological template, and the dark damage area type is determined by the one with the higher matching degree. Obtain information on hidden damage in leather texture and leather processing based on the type of hidden damage area.

4. The method for feature recognition of leather products based on a combination of computer vision and deep learning according to claim 3, characterized in that, The step of obtaining the light intensity attenuation region distribution based on the light intensity distribution matrix in the light transmission information includes: Obtain the light intensity value of each grid cell in the light intensity distribution matrix; The light intensity gradient value between adjacent grid cells is obtained based on the light intensity value of each grid cell. The light intensity gradient values ​​are thresholded and filtered, and the set of light intensity gradient values ​​greater than the threshold is generated into a set of abrupt gradient points; By expanding the region centered on the set of abrupt gradient points, a distribution of light intensity attenuation regions is generated.

5. The method for feature recognition of leather products based on a combination of computer vision and deep learning according to claim 4, characterized in that, The steps of obtaining the initial suspected dark damage area based on the light intensity attenuation region distribution, and obtaining the light intensity contrast matrix based on the initial suspected dark damage area, include: Establish a regularly distributed sampling grid within the initial suspected area of ​​hidden injury, and obtain multiple sampling points based on the sampling grid; With each sampling point as the center, multiple ring-shaped regions are constructed outwards to obtain the average light intensity value within each ring-shaped region; The degree of difference between the light intensity value of the central sampling point and the average light intensity value of each layer of the annular region is obtained, and the radial contrast sequence of the sampling point is obtained based on the degree of difference. The radial contrast sequence of each sampling point is normalized to obtain a normalized contrast vector. Based on the spatial location of each sampling point within the initial suspected dark injury area, the corresponding normalized contrast vectors are arranged into a two-dimensional matrix to form the light intensity contrast matrix.

6. The method for feature recognition of leather products based on a combination of computer vision and deep learning according to claim 5, characterized in that, The steps of obtaining leather texture hidden damage information and leather processing hidden damage information based on the type of hidden damage area include: Based on the type of the hidden injury area, the discrimination result information is obtained; When the determination result information is texture dark damage, the position coordinates of the dark damage area are obtained. Based on the position coordinates, the area parameter and the concavity and convexity parameter are extracted from the shape parameters of the dark damage area. The position coordinates, area parameter and concavity and convexity parameter are integrated to generate the leather texture dark damage information. When the judgment result information is a hidden damage in the leather processing, the position coordinates of the hidden damage area are obtained. Based on the position coordinates, the shape complexity parameter is extracted from the shape parameter of the hidden damage area. Based on the position coordinates, the light intensity attenuation amplitude is extracted from the light intensity contrast matrix. The position coordinates, shape complexity parameter and light intensity attenuation amplitude are integrated to generate the hidden damage information of the leather processing.

7. The method for feature recognition of leather products based on a combination of computer vision and deep learning according to claim 6, characterized in that, The step of obtaining the leather texture evaluation value based on the leather damage information and elastic deformation information includes: The number of hidden defects per unit area is obtained based on the position coordinates in the hidden defect information of the leather texture, and a hidden defect density factor is generated based on the number of hidden defects. The proportion of the dark damage area to the total area of ​​the detection area is obtained based on the area parameter in the dark damage information of the leather texture, and the dark damage area proportion factor is generated based on the proportion of the dark damage area to the total area of ​​the detection area. Based on the elastic deformation information, the displacement vector of the feature point is obtained, and the displacement vector of the feature point is mapped according to its original position on the surface texture image set to generate a displacement field distribution map. The deformation compatibility coefficient is obtained from the displacement field distribution diagram. By comparing the regions with high dark damage density factors with their corresponding texture response intensities, the deformation suppression index of the dark damage regions is obtained. Based on the distribution characteristics of the dark damage and the texture deformation characteristics, the texture-deformation correlation degree is obtained; Based on the texture-deformation correlation, obtain the texture stability index; The leather texture evaluation value is generated by weighted fusion based on the texture stability index and the dark damage distribution characteristics.

8. The method for feature recognition of leather products based on the combination of computer vision and deep learning according to claim 1, characterized in that, The step of generating a leather processing evaluation value based on the leather processing damage information and elastic deformation information includes: The spatial coordinate set of the hidden damage area is obtained based on the position coordinates in the leather processing hidden damage information; Generate a regional deformation response sequence based on the spatial coordinate set of the hidden injury area; The deformation gradient of the dark injury region is obtained based on the regional deformation response sequence. The heterogeneity of the defect structure is obtained based on the shape complexity parameter of the hidden injury area; The internal density variation of the defect is obtained based on the light intensity attenuation amplitude in the leather processing dark damage information. Based on the deformation gradient of the hidden damage area, the heterogeneity of the defect structure, and the density variation of the defect, a comprehensive impact index of processing defects is generated. The leather processing evaluation value is generated by mapping and matching the comprehensive impact index of processing defects with the preset processing quality benchmark spectrum.

9. The method for feature recognition of leather products based on a combination of computer vision and deep learning according to claim 8, characterized in that, The step of obtaining the comprehensive leather evaluation score based on the leather texture evaluation value and the leather processing evaluation value includes: The spatial coordinate set of the hidden damage area is obtained based on the position coordinates in the leather processing hidden damage information; Deformation response of multiple hidden damage regions based on the spatial coordinate set of the hidden damage region; Obtain the structural distortion index of multiple hidden injury areas; The light intensity attenuation level is divided based on the light intensity attenuation amplitude within the dark damage area. The area ratio of different attenuation level areas within the dark damage area is obtained, and the ratio of the boundary length between each attenuation level area to the total boundary length is obtained based on the area ratio. The density non-uniformity parameter is obtained based on the area ratio and the proportion, and the density variation characterization value of multiple dark injury areas is obtained based on the density non-uniformity parameter. The coupling influence factor of processing defects is obtained based on the deformation response degree; Obtain multiple quality grades and their corresponding ranges of coupling influence factors, and obtain the final leather processing evaluation value based on the ranges of coupling influence factors and the coupling influence factors of processing defects.

10. The method for feature recognition of leather products based on the combination of computer vision and deep learning according to claim 1, characterized in that, The step of generating leather assessment information based on the overall leather assessment degree includes: Based on the comprehensive leather evaluation score, the comprehensive leather evaluation score is compared with multiple preset quality grade threshold ranges to obtain the comparison results; The overall impact of the defect is obtained based on the comparison results. The overall impact of the defect is compared with a preset defect severity threshold, and the comparison result is obtained. The defect severity level is then obtained based on the comparison result. A defect severity classification is generated based on the aforementioned defect severity level; Based on the defect severity classification, spatial location information of the hidden damage area, and defect type information, a repair urgency index for multiple hidden damage areas is obtained; and the multiple hidden damage areas are sorted according to the repair urgency index to generate a defect repair priority sequence. Based on the quality grading identifier, the overall impact of defects, the classification of defect severity, and the priority sequence of defect repair, information is integrated to generate structured leather assessment information.

Images