A sole wear detection system based on image recognition

Abstract

The present invention discloses a sole wear detection system based on image recognition, relating to the field of image recognition technology. The system comprises: an image acquisition module, an image processing module, a wear analysis module, a database module, and a decision output module. The image acquisition module is used to acquire multi-angle high-definition images of the sole; the image processing module is used to remove impurities from the multi-angle high-definition images; the database module is used to store data on different shoe types; the wear analysis module distinguishes between normal wear and abnormal wear based on the database module and the three-dimensional distribution data; and the decision output module is used to output a quality assurance determination conclusion and a repair plan. The present invention provides a sole wear detection system based on image recognition. Through three-dimensional morphology reconstruction and a deep learning model, the system accurately distinguishes between natural wear and abnormal wear.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a sole wear detection system based on image recognition. Background Art

[0002] Existing work shoes and insulating shoes have various ridges on the soles of different sizes and depths. During use, some smaller and shallower ridges will wear out earlier. Therefore, a sole wear detection system is used to detect the wear status of the soles.

[0003] Existing sole wear detection relies primarily on manual visual inspection or simple thickness measurement, using calipers or handheld thickness gauges to compare thickness at a single point on a specific area of ​​the sole to determine the degree of wear. Some technologies use two-dimensional image scanning devices to capture flat images of the sole and identify worn areas based on grayscale changes, but these technologies are unable to resolve differences in three-dimensional topography.

[0004] Existing sole wear detection technology has the following defects: 1. Existing technology relies on manual visual inspection to distinguish between natural wear and human damage, which is easily affected by the experience of quality inspectors. The accuracy rate of identifying malicious complaints is low, and there is a lack of correlation analysis between wearing habits and wear patterns. It is difficult to identify malicious wear caused by short-term abnormal use, and cannot effectively detect sole tampering and counterfeiting, posing a risk of fraudulent claims; 2. Existing technology only compares local thickness data and cannot fully reflect the changes in the three-dimensional morphology of the sole, resulting in inaccurate maintenance plan formulation.

[0005] Therefore, we propose a sole wear detection system based on image recognition. Summary of the Invention

[0006] The main purpose of the present invention is to provide a sole wear detection system based on image recognition. Through this system, it is possible to solve the problems in the existing technology that rely on manual visual inspection to distinguish between natural wear and human damage, are easily affected by the experience of quality inspectors, have a low accuracy rate in identifying malicious complaints, lack correlation analysis between wearing habits and wear patterns, have difficulty in identifying malicious wear caused by short-term abnormal use, cannot effectively detect sole tampering and counterfeiting, and have the risk of fraudulent claims; moreover, the existing technology only compares local thickness data and cannot fully reflect the three-dimensional morphological changes of the sole, resulting in inaccurate maintenance plan formulation.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A sole wear detection system based on image recognition includes: an image acquisition module, an image processing module, a wear analysis module, a database module and a decision output module;

[0009] An image acquisition module, configured to acquire multi-angle high-definition images of the sole;

[0010] An image processing module is configured to remove impurities from the multi-angle high-definition images, generate a three-dimensional image model of the sole based on the multi-angle high-definition images through a preset three-dimensional reconstruction unit, and analyze the three-dimensional image model with preset standard three-dimensional topographic features of the sole through a preset processing unit to obtain three-dimensional distribution data of the wear area of ​​the sole;

[0011] A database module, the database module is used to store standard three-dimensional morphological features of the soles of different shoe types, user wearing history data, malicious wear feature library and standard wear data;

[0012] A wear analysis module, based on a pre-trained deep learning model, performs the following determinations: distinguishing between normal wear and abnormal wear based on the database module and the three-dimensional distribution data, wherein abnormal wear includes non-uniform deformation caused by sole quality issues and regular geometric wear caused by malicious user damage; and generating a quantitative wear score based on the normal wear and abnormal wear.

[0013] A decision output module is used to combine the quantitative wear score with a preset threshold value to output a quality assurance judgment conclusion and a maintenance plan.

[0014] Preferably, communication connections are established among the image acquisition module, image processing module, wear analysis module, database module and decision output module;

[0015] The image acquisition module includes a rotatable stage, a ring-shaped LED light source array and a high-resolution industrial camera.

[0016] Preferably, the image processing module performs impurity removal processing on the multi-angle high-definition image, and the specific steps include:

[0017] S1. Positionally calibrate the multi-angle HD images of the sole to eliminate image offsets caused by differences in shooting angles. Use color recognition and contour extraction techniques to separate the sole from the background in the multi-angle HD images. Remove interfering pixels from the rotatable stage, shadows, and non-sole areas, retaining multiple first complete sole images.

[0018] S2. Dynamically adjust the parameters of the annular LED light source array and the image brightness and darkness balance to address bright spots or mirror effects caused by light reflection in the plurality of first complete sole images, thereby eliminating interference of highlight areas with texture details. Furthermore, a preset restoration unit automatically identifies missing areas in the plurality of first complete sole images due to sole stains or shooting angle limitations, predicts and fills in the image content of the missing areas using texture features of adjacent areas, and generates a plurality of second complete sole images.

[0019] S3. Reduce the random noise in multiple second complete sole images through multi-level filtering, smooth the granular spots caused by environmental interference, enhance the contour features of the sole pattern edge and the worn area, highlight the excessive difference between the worn boundary and the normal area, and complete the de-noising processing of the multi-angle high-definition image of the sole.

[0020] Preferably, the three-dimensional distribution data of the sole wear area in the image processing module is obtained by the following steps:

[0021] Q1, the 3D reconstruction unit uses the plurality of second complete sole images that have undergone de-cluttering processing to extract spatial position information of the sole at different viewing angles, generates a set of 3D spatial data points of the sole surface through geometric calculation, and connects the spatial relationships between adjacent 3D spatial data points according to the distribution pattern of the 3D spatial data point set to form a continuous 3D image model of the sole;

[0022] Q2. Mirroring the three-dimensional image model and the standard three-dimensional topographic features of the sole is performed by the processing unit, and a key point alignment method is preset to eliminate spatial deviations caused by shooting angles. A preset reference datum plane is marked between the three-dimensional image model and the standard three-dimensional topographic features of the sole, and vertical distances between the three-dimensional image model and the reference datum plane and between the standard three-dimensional topographic features of the sole and the reference datum plane are calculated point by point to generate a first depth change value set and a second depth change value set;

[0023] Q3. Compare the first depth change value set with the second depth change value set, analyze the values ​​in the first depth change value set that exceed or are less than the values ​​in the second depth change value set, generate a first three-dimensional wear distribution map of depth differences, and generate a second three-dimensional wear distribution map including position and area ratio based on the first three-dimensional wear distribution map, and mark the wear parameters of key areas of the forefoot, heel and arch of the sole in the first three-dimensional wear distribution map and the second three-dimensional wear distribution map to obtain three-dimensional distribution data of the wear area of ​​the sole.

[0024] Preferably, the specific steps of generating the second three-dimensional wear distribution map including the position and area ratio according to the first three-dimensional wear distribution map in step Q3 are as follows:

[0025] Q31. Based on the depth difference data of the first three-dimensional wear distribution map, delineate areas on the surface of the three-dimensional image model where the depth variation exceeds a preset threshold, mark them as candidate wear areas, and record the center coordinates and boundary range of each candidate wear area;

[0026] Q32. For each candidate wear area, calculate the sole surface area covered based on its boundary range, and convert the ratio of the covered sole surface area to the overall sole surface area to obtain an area ratio of each candidate wear area;

[0027] Q33. Classify the candidate wear areas into three wear levels: light, moderate, and heavy, based on a combination of the depth difference data and the area percentage value of the first three-dimensional wear distribution map;

[0028] Q34. In the surface coordinate system of the three-dimensional image model, using the center coordinates of the wear candidate area as an anchor point, generate a circular coverage area, the radius of which is proportional to the area ratio, and the color depth of the circular coverage area is positively correlated with the wear level;

[0029] Q35. Mark the area ratio of each wear candidate area at the center of the corresponding circular coverage area, and associate the adjacent wear candidate areas through connecting lines to form a second three-dimensional wear distribution map showing the wear distribution density and severity.

[0030] Preferably, the specific steps for distinguishing normal wear from abnormal wear in the wear analysis module are:

[0031] H1, extracting the position distribution, shape regularity, and depth difference data of each candidate wear area from the second three-dimensional wear distribution map, and calculating the area proportions and depth extremes of key areas of the forefoot, heel, and arch of the sole;

[0032] H2. Compare the extracted wear characteristics of the candidate wear area with the standard wear data in the database module to analyze whether the candidate wear area conforms to the distribution law of human gait mechanics. If the wear is concentrated in the common force-bearing area and the shape is naturally diffused, mark it as normal wear;

[0033] H3: Abnormal wear is determined when at least one of the following conditions is detected:

[0034] a. The wear candidate area has regular geometric features and uniform depth variation, which is consistent with the tool damage characteristics;

[0035] b. Isolated wear areas with sudden changes in depth appear in the non-stressed areas of the sole;

[0036] c. The total area ratio of multiple wear areas on the same sole exceeds the normal threshold corresponding to the wearing time;

[0037] d. The texture similarity between the wear candidate area and the human damage cases in the malicious wear feature library exceeds 90%.

[0038] Preferably, the steps for generating the quantitative wear score in the wear analysis module are as follows:

[0039] F1. Based on the location, area percentage, and depth difference data of the candidate wear areas marked in the second three-dimensional wear distribution map, as well as the normal wear and abnormal wear, extract the cumulative area percentage, maximum depth value, number of abnormal wear areas, and distribution density parameters of each candidate wear area. Combined with the standard wear data of the same shoe model in the database module, dynamic weights are assigned to the wear parameters of different candidate wear areas, where the area percentage weights of the forefoot and heel areas of the sole are higher than those of the arch area, and the weight of the abnormal wear area is three times that of the normal wear area.

[0040] F2. Calling the user's wearing history data in the database module and adjusting the baseline threshold of the wear score according to the actual wearing time and usage scenario;

[0041] F3. If there are candidate wear areas with abnormal wear or normal wear, an additional score deduction value is added based on the degree of matching between their regular geometric features and the malicious wear feature library. The deduction value is proportional to the area ratio of the abnormal wear area and inversely proportional to the area ratio of the normal wear area.

[0042] F3. The wear parameters assigned with dynamic weights, the baseline threshold, and the deduction value are superimposed to generate a quantitative wear score of 0-100 points, which is divided into four levels: 80 points or more for slight wear, 60-80 points for moderate wear, 30-60 points for heavy wear, and 30 points or less for severe damage.

[0043] Preferably, the specific operation process of the decision output module is:

[0044] C1. Matching a preset maintenance plan library based on the level corresponding to the quantitative wear score, and extracting the maintenance cost corresponding to each level;

[0045] C2. Based on the abnormal wear type, the following warranty judgment conclusion is obtained:

[0046] If the abnormal wear is caused by non-uniform deformation due to sole quality problems, the output is "in compliance with the warranty range"; if the abnormal wear conforms to the regular geometric wear characteristics caused by malicious damage by the user, the output is "artificial damage"; if the quantitative wear score of the normal wear area is lower than the preset threshold but does not trigger an abnormality judgment, the output is "normal wear";

[0047] C3. Generate corresponding repair steps based on the repair cost and the warranty determination conclusion, integrate them into a visual report, mark the repair priority and cost details of each wear area, and output it to a preset terminal.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] First, this invention uses 3D morphology reconstruction and deep learning models to accurately distinguish between natural wear and abnormal wear. Combined with texture matching technology from a malicious wear signature library, it effectively identifies malicious tampering caused by vandalism or short-term abnormal use. Compared to manual visual inspection, the detection accuracy is increased to over 90%, significantly reducing the risk of fraud in insurance claims. It also supports sole quality traceability and provides merchants with objective evidence of malicious complaints.

[0050] Second, the present invention uses multi-angle 3D reconstruction technology to compare the depth difference between the actual sole and the standard model point by point on the reference plane, accurately capturing local thickness changes and overall deformation trends, avoiding the one-sidedness of traditional single thickness detection. Combined with dynamic weighted evaluation of multiple parameters such as the wear area ratio, depth extremes, and distribution density, a quantitative score and 3D distribution map are generated to provide accurate data support for maintenance plans, improve maintenance efficiency, and reduce the return rate.

[0051] 3. In this invention, the maintenance plan and the quality assurance conclusion are automatically matched based on the quantitative score, and a visual report is output, which reduces the cost of manual intervention and improves the quality inspection efficiency by 80%. It is suitable for the large-scale inspection needs of shoe companies and insurance institutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a system flow chart of the present invention;

[0053] Figure 2 This is a flow chart of the present invention for performing impurity removal processing on a multi-level high-definition image;

[0054] Figure 3 A flow chart of generating three-dimensional distribution data of the present invention;

[0055] Figure 4 Flowchart for the generation of a quantitative wear score for the present invention.

[0056] In the figure: 1. Image acquisition module; 2. Image processing module; 3. Wear analysis module; 4. Database module; 5. Decision output module. DETAILED DESCRIPTION

[0057] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0058] like Figure 1-4 As shown, a sole wear detection system based on image recognition includes: an image acquisition module 1, an image processing module 2, a wear analysis module 3, a database module 4 and a decision output module 5;

[0059] Image acquisition module 1 is used to capture multi-angle, high-definition images of the shoe sole. It includes a rotatable stage, a ring-shaped LED light source array, and a high-resolution industrial camera. The rotatable stage automatically adjusts the shoe's angle, while the ring-shaped LED light source dynamically adjusts the light to eliminate reflective interference. The high-resolution industrial camera simultaneously captures multi-angle, high-definition images. The stage precisely rotates to ensure 360-degree coverage, while the light source intelligently adapts to the sole material to balance brightness and color temperature. The high-resolution industrial camera captures details at a resolution of 20 megapixels. These three elements work together to achieve seamless, high-definition image capture, providing standardized data for subsequent 3D reconstruction. Multi-angle capture of a single shoe is completed within 10 seconds, and the camera is compatible with soles of different materials and colors.

[0060] Image processing module 2 is used to remove impurities from the multi-angle high-definition images, generate a three-dimensional image model of the sole based on the multi-angle high-definition images through a preset three-dimensional reconstruction unit, and analyze the three-dimensional image model with preset standard three-dimensional topographic features of the sole through a preset processing unit to obtain three-dimensional distribution data of the sole wear area;

[0061] The image processing module 2 performs impurity removal processing on the multi-angle high-definition image, and the specific steps include:

[0062] S1. Positionally calibrate the multi-angle HD images of the sole to eliminate image offsets caused by differences in shooting angles. Use color recognition and contour extraction techniques to separate the sole from the background in the multi-angle HD images. Remove interfering pixels from the rotatable stage, shadows, and non-sole areas, retaining multiple first complete sole images.

[0063] The position calibration and background separation of multi-angle sole images are achieved through the following process: First, based on the rotation angle parameters of the rotatable stage, preset markers (such as the positioning marks on the edge of the stage) are located in each sole image. The images taken at different angles are aligned to a unified spatial reference through a coordinate system conversion algorithm to eliminate the perspective offset caused by rotation. Then, the color gamut difference between the sole and the background (such as the sole is dark rubber and the stage is light matte) is used to extract the main contour of the sole through HSV color space threshold segmentation. The sole boundary is strengthened by combining edge detection technology, and interfering areas such as the edge of the stage, equipment reflections, and projected shadows are intelligently identified and eliminated. Finally, a pixel-level masking operation is performed to retain only the main area of ​​the sole to generate a pure image without background interference, ensuring the accuracy of subsequent 3D reconstruction.

[0064] S2. Dynamically adjust the parameters of the annular LED light source array and the image brightness and darkness balance to address bright spots or mirror effects caused by light reflection in the multiple first complete sole images, eliminating interference of highlight areas on texture details. A preset restoration unit automatically identifies missing areas in the multiple first complete sole images due to sole stains or shooting angle limitations, predicts and fills in the image content of the missing areas using texture features of adjacent areas, and generates multiple second complete sole images.

[0065] In step S2, the interference of the highlight area on the texture details is eliminated, and the specific contents include:

[0066] S21, automatically identifying overly bright areas caused by reflections in the first complete sole image, and determining the interference range based on brightness and color characteristics;

[0067] S22. Dynamically adjust the illumination angle and intensity of the ring-shaped LED light array, reshoot or adjust lighting conditions for highlight areas to reduce overexposure;

[0068] S23. Combining the clear image captured after adjusting the light with the first complete sole image, and covering the highlight portion of the first complete sole image with the details of the normal area to ensure the integrity of the texture;

[0069] The steps for generating the second complete sole image in step S2 are as follows:

[0070] S24, detecting pattern breakage and color discontinuity caused by stains or occlusion in the first complete sole image, and marking the missing areas;

[0071] S25. Analyze the pattern direction, particle density and other characteristics around the missing area to infer the pattern and structure of the missing part;

[0072] S26, generating a filling content consistent with the surrounding area based on the inference result and smoothly connecting the boundaries, so that the repaired area and the first complete sole image are naturally integrated, thereby generating a second complete sole image;

[0073] S3. Reduce random noise in multiple second complete sole images through multi-level filtering, smooth out granular spots caused by environmental interference, enhance the contour features of the sole pattern edges and worn areas, and highlight the excessive difference between the worn boundaries and normal areas, thus completing the de-noising process of the multi-angle high-definition images of the soles;

[0074] The removal of the multi-angle high-definition image of the sole in step S3 includes the following steps:

[0075] S31, random noise removal: performing preliminary filtering on the second complete sole image, reducing the brightness differences of isolated pixels in the image layer by layer, thereby reducing small noise caused by environmental interference, while preserving the main structure of the sole pattern;

[0076] S32, Grain and Spot Smoothing: In the image after preliminary noise reduction, we use local blurring and edge preservation techniques to gradually reduce the light and dark contrast of the granular areas caused by dust or uneven material on the sole surface, thereby smoothing the surface texture transition.

[0077] S33, Contour Feature Enhancement: Adaptively adjust the contrast of the processed image to enhance the light and dark differences at the edge of the sole pattern. Use boundary detection technology to highlight the transition line between the worn area and the normal area, making the worn contour clearly visible, and ultimately output a high-precision cleaned image.

[0078] Example 1: A pair of sneakers is mounted on a rotatable stage. After the system is activated, the stage automatically rotates eight times at 45° intervals. The circular LED light array dynamically adjusts to warm light mode based on the sole material, while a high-resolution industrial camera simultaneously captures 20-megapixel HD images. The image processing module aligns images at various angles through coordinate system transformation, extracts the sole body using HSV color space segmentation, and removes background interference. For reflective areas in the forefoot, the system automatically reduces the brightness of the corresponding LED subarea and re-images the image, replacing the highlights in the original image. The system then detects missing patterns on the arch of the sole due to stains, generates fill-in content based on adjacent textures, and outputs a high-precision, cleaned image after multi-level noise reduction and contour enhancement.

[0079] The three-dimensional distribution data of the shoe sole wear area in the image processing module 2 is obtained by the following steps:

[0080] Q1. The 3D reconstruction unit uses the plurality of second complete sole images that have undergone de-cluttering processing to extract spatial position information of the sole from different viewing angles, generates a set of 3D spatial data points of the sole surface through geometric calculation, and connects the spatial relationships between adjacent 3D spatial data points according to the distribution pattern of the 3D spatial data point set to form a continuous 3D image model of the sole;

[0081] In step Q1, a three-dimensional spatial data point set of the sole surface is generated through geometric calculations. Specifically, the following steps are performed: first, based on the position offset of the same sole feature point (such as a pattern intersection) in multiple second complete sole images at different viewing angles, a spatial geometric model is constructed in combination with camera parameters (focal length, shooting angle), and the three-dimensional coordinates of each feature point are calculated; then, based on the relative distance and angle relationship between adjacent feature points, the surface curvature of the area not directly photographed is inferred, and a dense data point set covering the entire sole is generated. Finally, a continuous three-dimensional surface is formed through point cloud interpolation;

[0082] Q2. A processing unit performs mirror processing on the three-dimensional image model and the standard three-dimensional topographic features of the sole, and presets a key point alignment method to eliminate spatial deviation caused by the shooting angle, marks a preset reference reference plane between the three-dimensional image model and the standard three-dimensional topographic features of the sole, calculates the vertical distance between the three-dimensional image model and the reference reference plane, and between the standard three-dimensional topographic features of the sole and the reference reference plane, point by point, to generate a first depth change value set and a second depth change value set;

[0083] Among them, the preset key point alignment method in step Q2 eliminates the spatial deviation caused by the shooting angle. The steps are as follows: at least five groups of corresponding feature points are selected at the forefoot, heel center and pattern nodes of the sole of the 3D image model and the standard 3D morphological features of the sole as alignment references; the coordinate position of the 3D image model is adjusted through rotation and translation operations to minimize the spatial position deviation of the two groups of feature points and eliminate the spatial offset caused by the difference in shooting angle; then, based on the consistency of the surface curvature of the 3D image model and the standard 3D morphological features of the sole, the alignment accuracy is further optimized until the overall shape matches;

[0084] Step Q2 calculates point-by-point the vertical distances between the 3D image model and the reference plane, and between the standard 3D topography of the sole and the reference plane. This includes setting a horizontal reference plane between the 3D image model and the standard 3D topography of the sole, calculating the shortest vertical distance between each point on the surface of the 3D image model and the reference plane, and calculating the vertical distance between each point on the surface of the standard 3D topography of the sole and the reference plane, thereby generating two sets of depth values ​​for quantifying the deformation difference between the actual sole and the standard 3D topography of the sole.

[0085] Q3. Compare the first depth change value set with the second depth change value set, analyze the values ​​in the first depth change value set that exceed or are less than the values ​​in the second depth change value set, generate a first three-dimensional wear distribution map of depth differences, and generate a second three-dimensional wear distribution map including position and area ratio based on the first three-dimensional wear distribution map, and mark the wear parameters of the key areas of the forefoot, heel and arch of the sole in the first three-dimensional wear distribution map and the second three-dimensional wear distribution map to obtain three-dimensional distribution data of the wear area of ​​the sole.

[0086] The specific steps of generating the second three-dimensional wear distribution map including the position and area ratio according to the first three-dimensional wear distribution map in step Q3 are as follows:

[0087] Q31. Based on the depth difference data of the first 3D wear distribution map, delineate areas on the surface of the 3D image model where the depth variation exceeds a preset threshold, mark them as candidate wear areas, and record the center coordinates and boundary range of each candidate wear area;

[0088] Q32. For each candidate wear area, calculate the sole surface area covered based on its boundary range, and then convert the ratio of the covered sole surface area to the overall sole surface area to obtain the area ratio of each candidate wear area.

[0089] The area percentage of each candidate wear area is calculated by dividing the surface of the 3D image model of the sole into tiny grid cells, calculating the actual area of ​​each grid cell, and summing up the areas of all grid cells within the boundary of the candidate wear area to obtain the surface area of ​​the candidate area. The sum of the areas of each candidate area is then converted into a percentage of the overall surface area of ​​the sole (i.e., the sum of the areas of all grid cells) to obtain the area percentage of each area.

[0090] Q33. Classify the candidate wear areas into three wear levels: light, moderate, and heavy, based on the combination of the depth difference data and the area ratio value of the first three-dimensional wear distribution map;

[0091] Q34. In the surface coordinate system of the 3D image model, with the center coordinates of the wear candidate area as the anchor point, a circular coverage area is generated. The radius of the circular coverage area is proportional to the area ratio, and the color depth of the circular coverage area is positively correlated with the wear level.

[0092] Q35. Mark the area ratio of each wear candidate area at the center of the corresponding circular coverage area, and connect the adjacent wear candidate areas through connecting lines to form a second three-dimensional wear distribution map showing the wear distribution density and severity.

[0093] Example 2: After a pair of hiking shoes is photographed by the image acquisition module, the image processing module 2 performs the following process:

[0094] 3D reconstruction: Extracting features of the arch groove, forefoot anti-slip pattern, and heel support point from the cleaned multi-angle images, generating a 3D model containing 280,000 data points.

[0095] Key point alignment: Select the arch center depression, the first metatarsal node corresponding to the forefoot pattern, the heel center point, and two pattern intersections as alignment benchmarks, and adjust the model position so that the overall coordinate deviation is ≤0.25mm;

[0096] Depth comparison: Using the plane 1.2mm below the lowest point of the sole as the reference plane, the following depth differences were detected: 1.1mm in the forefoot area (standard 0.6mm); 0.8mm in the heel area (standard 0.4mm); and 0.4mm in the arch area (standard 0.2mm).

[0097] - Wear distribution generation: Forefoot demarcated 130mm above threshold (0.8mm) 2 area, accounting for the total surface area of ​​the sole (1150mm 2) 11.3%; 50mm of the heel is defined as exceeding the threshold (0.6mm) 2 area, accounting for 4.3%; the arch of the foot is defined as 35mm above the threshold (0.3mm) 2 area, accounting for 3.0%; output results: the wear coordinates, area proportion (11.3%, 4.3%, 3.0%) and depth difference data of the forefoot, heel and arch are marked in the 3D model.

[0098] Database module 4 is used to store standard 3D topographic features of the soles of different shoe types, user wearing history data, a malicious wear feature library, and standard wear data. Database module 4 uses a separate database and table storage method: 3D topographic features are stored in a dedicated model library in a point cloud encrypted format; user wearing history data is stored in a relational database in time series; the malicious wear feature library and standard wear data are stored in a structured table in an associated manner, and a feature vector index is established to support fast comparison and retrieval;

[0099] Wear analysis module 3, based on a pre-trained deep learning model, performs the following determinations: Based on database module 4 and three-dimensional distribution data, it distinguishes between normal wear and abnormal wear. Abnormal wear includes non-uniform deformation caused by sole quality issues and regular geometric wear caused by malicious user damage. It also generates a quantitative wear score based on normal wear and abnormal wear.

[0100] The specific steps for distinguishing normal wear from abnormal wear in the wear analysis module 3 are as follows:

[0101] H1. Extract the location distribution, shape regularity, and depth difference data of each wear candidate area from the second 3D wear distribution map, and calculate the area proportion and depth extremes of the key areas of the forefoot, heel, and arch of the sole;

[0102] H2. Compare the wear characteristics of the extracted candidate wear areas with the standard wear data in database module 4 to analyze whether the candidate wear areas conform to the distribution law of human gait mechanics. If the wear is concentrated in the common stress area and the shape is naturally diffused, it is marked as normal wear;

[0103] H3: Abnormal wear is determined when at least one of the following conditions is detected:

[0104] a. The wear candidate area has regular geometric features and uniform depth variation, which is consistent with the tool damage characteristics;

[0105] b. Isolated wear areas with sudden changes in depth appear in the non-stressed areas of the sole;

[0106] c. The total area ratio of multiple wear areas on the same sole exceeds the normal threshold corresponding to the wearing time;

[0107] d. The texture similarity between the wear candidate area and the human damage cases in the malicious wear feature library exceeds 90%.

[0108] The steps for generating the quantitative wear score in the wear analysis module 3 are as follows:

[0109] F1. Based on the location, area percentage, and depth difference data of the wear candidate areas marked in the second three-dimensional wear distribution map, as well as the normal wear and abnormal wear data, extract the cumulative area percentage, maximum depth value, number of abnormal wear areas, and distribution density parameters of each wear candidate area. Combined with the standard wear data of the same shoe model in database module 4, dynamic weights are assigned to the wear parameters of different wear candidate areas. The area percentage weights of the forefoot and heel areas of the sole are higher than those of the arch area, and the weight of the abnormal wear area is three times that of the normal wear area.

[0110] F2. Call the user's wearing history data in database module 4 and adjust the baseline threshold of the wear score according to the actual wearing time and usage scenario;

[0111] F3. If there are candidate wear areas with abnormal wear or normal wear, an additional score deduction value will be added based on the matching degree of its regular geometric features with the malicious wear feature library. The deduction value is proportional to the area ratio of the abnormal wear area and inversely proportional to the area ratio of the normal wear area.

[0112] F3. The wear parameters, baseline thresholds, and deduction values ​​that have been assigned dynamic weights are superimposed to generate a quantitative wear score of 0-100 points, which is divided into four levels: 80 points or more for slight wear, 60-80 points for moderate wear, 30-60 points for heavy wear, and below 30 points for severe damage.

[0113] The decision output module 5 is used to combine the quantitative wear score with the preset threshold value to output the quality assurance judgment conclusion and maintenance plan.

[0114] Among them, communication connections are established among the image acquisition module 1 , the image processing module 2 , the wear analysis module 3 , the database module 4 and the decision output module 5 .

[0115] The specific operation process of the decision output module 5 is as follows:

[0116] C1. Based on the level corresponding to the quantitative wear score, match the preset maintenance plan library and extract the corresponding maintenance cost for each level;

[0117] C2. Based on the abnormal wear type, the following warranty judgment conclusions are obtained:

[0118] If the abnormal wear is caused by non-uniform deformation due to sole quality problems, the output is "in compliance with the warranty range"; if the abnormal wear conforms to the regular geometric wear characteristics caused by malicious damage by the user, the output is "human damage"; if the quantitative wear score of the normal wear area is lower than the preset threshold but does not trigger an abnormal judgment, the output is "normal wear";

[0119] C3. Generate corresponding repair steps based on the repair cost and warranty judgment conclusion, integrate them into a visual report, mark the repair priority and cost details of each wear area, and output it to the preset terminal.

[0120] Example 3: After a pair of sports shoes of a certain brand is tested by the system, the wear analysis module 3 and the decision output module 5 execute the following process:

[0121] Feature extraction: Parameters of the forefoot, heel, and arch wear areas were extracted from the three-dimensional distribution data. The forefoot area accounted for 12% with a maximum depth difference of 1.5mm, the heel accounted for 6% with a depth difference of 0.8mm, and the arch accounted for 2% with a depth difference of 0.3mm. A regular circular wear area (8mm in diameter and 1.2mm in depth) was found on the heel.

[0122] Abnormal determination: The regular circular wear area matches the "repeated scratching by hard objects" case in the malicious wear database with a 92% match, indicating human damage. The presence of isolated wear points (depth difference of 0.5mm) in the non-stress-bearing area of ​​the arch triggers an abnormal condition.

[0123] Scoring calculation: Normal wear weight distribution: forefoot (12% × 1.2), heel (6% × 1.0), arch (2% × 0.5); abnormal area deduction: regular wear area (0.5% area × 3 times weight) + isolated arch point (0.1% × 3 times), total deduction of 6 points; combined with the user's wear time (8 months), the baseline threshold is adjusted to 70 points, and the final score = 78 (normal) - 6 (abnormal) = 72 points (moderate wear);

[0124] Decision output: Warranty conclusion: Regular wear on the heel is considered "man-made damage", and isolated points on the arch are considered "quality defects", which partially meet the warranty requirements;

[0125] Repair plan: Forefoot patch repair (￥80), heel replacement (￥150, at your own expense), arch reinforcement (free of charge under warranty), and generate a visual report that will be pushed to the merchant terminal.

[0126] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A sole wear detection system based on image recognition, characterized in that: include: Image acquisition module, image processing module, wear analysis module, database module and decision output module; An image acquisition module, configured to acquire multi-angle high-definition images of the sole; An image processing module is configured to remove impurities from the multi-angle high-definition images, generate a three-dimensional image model of the sole based on the multi-angle high-definition images through a preset three-dimensional reconstruction unit, and analyze the three-dimensional image model with preset standard three-dimensional topographic features of the sole through a preset processing unit to obtain three-dimensional distribution data of the wear area of ​​the sole; A database module, the database module is used to store standard three-dimensional morphological features of the soles of different shoe types, user wearing history data, malicious wear feature library and standard wear data; A wear analysis module, based on a pre-trained deep learning model, performs the following determinations: distinguishing between normal wear and abnormal wear based on the database module and the three-dimensional distribution data, wherein abnormal wear includes non-uniform deformation caused by sole quality issues and regular geometric wear caused by malicious user damage; and generating a quantitative wear score based on the normal wear and abnormal wear. A decision output module, the decision output module is used to combine the quantitative wear score with a preset threshold value to output a quality assurance judgment conclusion and a maintenance plan; The three-dimensional distribution data of the sole wear area in the image processing module is obtained by the following steps: Q1, the 3D reconstruction unit uses the plurality of second complete sole images that have been processed for de-cluttering to extract spatial position information of the sole at different viewing angles, generates a set of 3D spatial data points of the sole surface through geometric calculation, and connects the spatial relationships between adjacent 3D spatial data points according to the distribution pattern of the 3D spatial data point set to form a continuous 3D image model of the sole; Q2. Mirroring the three-dimensional image model and the standard three-dimensional topographic features of the sole is performed by the processing unit, and a key point alignment method is preset to eliminate spatial deviations caused by shooting angles. A preset reference datum plane is marked between the three-dimensional image model and the standard three-dimensional topographic features of the sole, and vertical distances between the three-dimensional image model and the reference datum plane and between the standard three-dimensional topographic features of the sole and the reference datum plane are calculated point by point to generate a first depth change value set and a second depth change value set; Q3. Compare the first depth change value set with the second depth change value set, analyze the values ​​in the first depth change value set that exceed or are less than the values ​​in the second depth change value set, generate a first three-dimensional wear distribution map of depth differences, and generate a second three-dimensional wear distribution map including position and area ratio based on the first three-dimensional wear distribution map, and mark the wear parameters of key areas of the forefoot, heel and arch of the sole in the first three-dimensional wear distribution map and the second three-dimensional wear distribution map to obtain three-dimensional distribution data of the wear area of ​​the sole.

2. The shoe sole wear detection system based on image recognition according to claim 1, characterized in that: Communication connections are established between the image acquisition module, image processing module, wear analysis module, database module and decision output module; The image acquisition module includes a rotatable stage, a ring-shaped LED light source array and a high-resolution industrial camera.

3. The shoe sole wear detection system based on image recognition according to claim 2, characterized in that: The image processing module performs impurity removal processing on the multi-angle high-definition image, and the specific steps include: S1. Positionally calibrate the multi-angle HD images of the sole to eliminate image offsets caused by differences in shooting angles. Use color recognition and contour extraction techniques to separate the sole from the background in the multi-angle HD images. Remove interfering pixels from the rotatable stage, shadows, and non-sole areas, retaining multiple first complete sole images. S2. Dynamically adjust the parameters of the annular LED light source array and the image brightness and darkness balance to address bright spots or mirror effects caused by light reflection in the plurality of first complete sole images, thereby eliminating interference of highlight areas with texture details. Furthermore, a preset restoration unit automatically identifies missing areas in the plurality of first complete sole images due to sole stains or shooting angle limitations, predicts and fills in the image content of the missing areas using texture features of adjacent areas, and generates a plurality of second complete sole images. S3. Reduce the random noise in multiple second complete sole images through multi-level filtering, smooth the granular spots caused by environmental interference, enhance the contour features of the sole pattern edge and the worn area, highlight the excessive difference between the worn boundary and the normal area, and complete the de-noising processing of the multi-angle high-definition image of the sole.

4. The shoe sole wear detection system based on image recognition according to claim 3, characterized in that: The specific steps of generating the second three-dimensional wear distribution map including the position and area ratio according to the first three-dimensional wear distribution map in step Q3 are as follows: Q31. Based on the depth difference data of the first three-dimensional wear distribution map, delineate areas on the surface of the three-dimensional image model where the depth variation exceeds a preset threshold, mark them as candidate wear areas, and record the center coordinates and boundary range of each candidate wear area; Q32. For each candidate wear area, calculate the sole surface area covered based on its boundary range, and convert the ratio of the covered sole surface area to the overall sole surface area to obtain an area ratio of each candidate wear area; Q33. Classify the candidate wear areas into three wear levels: light, moderate, and heavy, based on a combination of the depth difference data and the area percentage value of the first three-dimensional wear distribution map; Q34. In the surface coordinate system of the three-dimensional image model, using the center coordinates of the wear candidate area as an anchor point, generate a circular coverage area, the radius of which is proportional to the area ratio, and the color depth of the circular coverage area is positively correlated with the wear level; Q35. Mark the area ratio of each wear candidate area at the center of the corresponding circular coverage area, and associate the adjacent wear candidate areas through connecting lines to form a second three-dimensional wear distribution map showing the wear distribution density and severity.

5. The shoe sole wear detection system based on image recognition according to claim 4, characterized in that: The specific steps for distinguishing normal wear from abnormal wear in the wear analysis module are: H1, extracting the position distribution, shape regularity, and depth difference data of each candidate wear area from the second three-dimensional wear distribution map, and calculating the area proportions and depth extremes of key areas of the forefoot, heel, and arch of the sole; H2. Compare the extracted wear characteristics of the candidate wear area with the standard wear data in the database module to analyze whether the candidate wear area conforms to the distribution law of human gait mechanics. If the wear is concentrated in the common force-bearing area and the shape is naturally diffused, mark it as normal wear; H3: Abnormal wear is determined when at least one of the following conditions is detected: a. The candidate wear area has regular geometric features and uniform depth variation, which is consistent with the tool damage characteristics; b. Isolated wear areas with sudden changes in depth appear in the non-stressed areas of the sole; c. The total area ratio of multiple wear areas on the same sole exceeds the normal threshold corresponding to the wearing time; d. The texture similarity between the wear candidate area and the human damage cases in the malicious wear feature library exceeds 90%.

6. The shoe sole wear detection system based on image recognition according to claim 5, characterized in that: The steps for generating the quantitative wear score in the wear analysis module are as follows: F1. Based on the location, area percentage, and depth difference data of the candidate wear areas marked in the second three-dimensional wear distribution map, as well as the normal wear and abnormal wear, extract the cumulative area percentage, maximum depth value, number of abnormal wear areas, and distribution density parameters of each candidate wear area. Combined with the standard wear data of the same shoe model in the database module, dynamic weights are assigned to the wear parameters of different candidate wear areas, where the area percentage weights of the forefoot and heel areas of the sole are higher than those of the arch area, and the weight of the abnormal wear area is three times that of the normal wear area. F2. Calling the user's wearing history data in the database module and adjusting the baseline threshold of the wear score according to the actual wearing time and usage scenario; F3. If there are candidate wear areas with abnormal wear or normal wear, an additional score deduction value is added based on the degree of matching between their regular geometric features and the malicious wear feature library. The deduction value is proportional to the area ratio of the abnormal wear area and inversely proportional to the area ratio of the normal wear area. F3. The wear parameters assigned with dynamic weights, the baseline threshold, and the deduction value are superimposed to generate a quantitative wear score of 0-100 points, which is divided into four levels: 80 points or more for slight wear, 60-80 points for moderate wear, 30-60 points for heavy wear, and 30 points or less for severe damage.

7. The shoe sole wear detection system based on image recognition according to claim 6, characterized in that: The specific operation process of the decision output module is as follows: C1. Matching a preset maintenance plan library based on the level corresponding to the quantitative wear score, and extracting the maintenance cost corresponding to each level; C2. Based on the abnormal wear type, the following warranty judgment conclusions are obtained: If the abnormal wear is caused by non-uniform deformation due to sole quality problems, the "warranty compliance" conclusion is output; if the abnormal wear conforms to the regular geometric wear characteristics caused by malicious damage by the user, the "human damage" conclusion is output; if the quantitative wear score of the normal wear area is lower than the preset threshold but does not trigger an abnormality judgment, the "normal wear" conclusion is output; C3. Generate corresponding repair steps based on the repair cost and the warranty determination conclusion, integrate them into a visual report, mark the repair priority and cost details of each wear area, and output it to a preset terminal.

Images