Vamp contour detection method and system based on image recognition model

Abstract

The invention relates to the technical field of computer vision, in particular to a vamp contour detection method and system based on an image recognition model. The method comprises the following steps: collecting original multi-state images of a vamp under different bending and extrusion degrees, and carrying out contour enhancement processing to obtain a deformation state contour set; key parts of the deformation state contour set are identified and marked, and a part marking graph is obtained; accurately positioning a part interface according to the part marking graph to obtain an accurate interface graph; constructing a three-dimensional hierarchical structure map according to the accurate interface map; performing local key point deformation tracking according to the three-dimensional hierarchical structure map to obtain multi-dimensional deformation parameters; and performing key part deformation analysis according to the multi-dimensional deformation parameter set to obtain a key point deformation characteristic table. Vamp deformation prediction under different conditions is achieved, the performance, comfort and durability of the product are improved, the development cost is reduced, and the development period is shortened.

Description

Technical Field

[0001] The present invention relates to the field of computer vision technology, and particularly to a method and system for detecting the upper contour based on an image recognition model. Background Art

[0002] Existing technologies usually only focus on the performance of the upper in a few specific states (such as completely flat, bent at a specific angle), and it is difficult to comprehensively reflect the real behavior of the upper during continuous deformation. The lack of fine analysis of a large number of intermediate states results in the inability to accurately capture the gradual change process and key turning points of the upper performance. In addition, the data between different states are often isolated, lacking effective association and integration, and it is difficult to form an overall understanding of the deformation behavior of the upper.

[0003] The upper is usually composed of multiple layers of materials (such as tongue, vamp, sole, lining, etc.), and the materials and structures of different layers contribute differently to the overall performance of the upper. However, most existing technologies analyze the upper as a whole, ignoring the structural differences and interactions between different layers. Even if some methods attempt to perform hierarchical analysis, they often stay at a rough regional division, making it difficult to accurately identify the boundaries of each layer, especially the interface between different layer materials, and even more unable to accurately analyze the relative displacement and cooperative relationship of each layer during deformation.

[0004] The different layers of the upper do not exist independently, but are tightly connected by stitching, gluing, etc. and affect each other during deformation. The interlayer contact, friction, stress transfer and other effects have an important impact on the overall deformation behavior of the upper. However, existing technologies rarely focus on the analysis of interlayer effects, lacking effective means to identify interlayer contact areas, track stress transfer paths, and quantify interlayer coupling effects. This leads to the inability to deeply understand the internal mechanism of upper deformation and makes it difficult to finely optimize the upper design.

[0005] In summary, there are problems in the existing technologies that it is impossible to comprehensively capture the continuous deformation process of the upper, difficult to finely analyze the multi-layer structure of the upper, and lack of in-depth research on interlayer interactions, which need to be solved urgently. Summary of the Invention

[0006] Based on this, it is necessary to provide a method and system for detecting the upper contour based on an image recognition model to solve at least one of the above technical problems.

[0007] To achieve the above object, a method for detecting the upper contour based on an image recognition model includes the following steps:

[0008] Step S1: Collect the original multi-state images of the upper under different bending and extrusion degrees, and perform contour enhancement processing to obtain a set of deformed state contours;

[0009] Step S2: Identify and mark the key parts of the deformation state contour set to obtain a part marking map; perform precise positioning of the part interface according to the part marking map to obtain a precise interface map; construct a three-dimensional hierarchical structure map according to the precise interface map;

[0010] Step S3: Perform local key point deformation tracking according to the three-dimensional hierarchical structure map to obtain multi-dimensional deformation parameters; perform key part deformation analysis according to the multi-dimensional deformation parameter set to obtain a key point deformation characteristic table;

[0011] Step S4: Identify the interlayer contact area distribution map according to the key point deformation characteristic table and the three-dimensional hierarchical structure map; perform analysis of the interlayer stress transfer path for the interlayer contact area distribution map to obtain a stress transfer path map; calculate the interlayer coupling coefficient according to the stress transfer path map and the key point deformation characteristic table to obtain an interlayer coupling coefficient matrix; perform interlayer deformation response modeling according to the interlayer coupling coefficient matrix and the interlayer contact area distribution map to obtain an interlayer interaction deformation model.

[0012] The present invention can obtain a large amount of multi-dimensional and high-quality original data on the deformation of the shoe upper by constructing a high-precision shoe upper deformation test platform and a multi-view image acquisition system. Through fine image preprocessing, segmentation, and contour extraction, a set of deformation state contours that precisely correspond to the deformation parameters is obtained. This provides a reliable data basis for subsequent structural analysis, deformation tracking, and modeling, avoiding the subjective errors and low efficiency of manual measurement, and realizing the automatic and high-precision acquisition and feature expression of the shoe upper deformation data. By combining texture features, clustering algorithms, prior knowledge, and precise boundary positioning techniques, the automatic identification, marking, and interface extraction of the key parts of the shoe upper (tongue, upper, sole) are realized. The constructed three-dimensional hierarchical structure map not only contains the geometric information (boundary coordinates) of each part of the shoe upper but also integrates structural information such as hierarchical relationships, connection relationships, and connection strengths. This provides structural constraints for subsequent deformation analysis, making deformation tracking and stress analysis more targeted and accurate. At the same time, it also provides an important structural model for the digital design and virtual simulation of footwear products. By defining and tracking the deformation trajectories of the key points of the shoe upper and combining local feature analysis and non-rigid deformation field modeling, multi-dimensional deformation parameters (displacement, curvature, strain, deformation rate) of the shoe upper in different deformation states can be obtained. Through in-depth analysis of the deformation characteristics (bending characteristics, folding patterns, stability) of the key parts (toe, tongue, heel), the internal mechanisms and laws of the shoe upper deformation are revealed. This provides quantitative indicators and scientific bases for the performance evaluation, comfort analysis, and optimization design of footwear products, helping to improve the wearing experience and functionality of footwear products. By identifying the interlayer contact area, analyzing the stress transfer path, calculating the interlayer coupling coefficient, and constructing the interlayer constraint equation, the interaction between different layers of the shoe upper (tongue, upper, sole) is deeply studied. The established interlayer interaction deformation model can accurately predict the overall deformation response of the shoe upper under different bending and extrusion conditions, including the displacement, stress, strain distribution of each layer, and the change of the overall contour. This provides a powerful tool for the structural design, material selection, and process optimization of footwear products, enabling effective prediction and control of the shoe upper deformation, improving the performance, comfort, and durability of the products, and reducing the development cost and cycle. Therefore, the present invention provides a method for detecting the shoe upper contour based on an image recognition model, which captures the continuous deformation process through multi-state acquisition, realizes fine part segmentation and interface positioning through structural hierarchy separation, reveals the interlayer interaction mechanism through interlayer interaction modeling, and finally establishes an interlayer interaction deformation model to predict the shoe upper deformation under any conditions, overcoming the deficiencies of the prior art in multi-state analysis, hierarchical analysis, and interlayer interaction analysis.

[0013] Preferably, the present invention also provides a shoe upper contour detection system based on an image recognition model for performing the method for detecting the shoe upper contour based on an image recognition model as described above. The shoe upper contour detection system based on an image recognition model includes:

[0014] A multi-state contour acquisition module for acquiring the original multi-state images of the shoe upper under different bending and extrusion degrees, and performing contour enhancement processing to obtain a set of deformation state contours;

[0015] A structural hierarchy separation module for identifying and marking the key parts of the set of deformation state contours to obtain a part marking map; accurately positioning the part interface according to the part marking map to obtain an accurate interface map; constructing a three-dimensional hierarchical structure map according to the accurate interface map;

[0016] A key point deformation tracking module for performing local key point deformation tracking according to the three-dimensional hierarchical structure map to obtain multi-dimensional deformation parameters; performing key part deformation analysis according to the multi-dimensional deformation parameter set to obtain a key point deformation characteristic table;

[0017] An interlayer interaction response modeling module for identifying the distribution map of interlayer contact areas according to the key point deformation characteristic table and the three-dimensional hierarchical structure map; analyzing the interlayer stress transfer path of the interlayer contact area distribution map to obtain a stress transfer path map; calculating the interlayer coupling coefficient according to the stress transfer path map and the key point deformation characteristic table to obtain an interlayer coupling coefficient matrix; performing interlayer deformation response modeling according to the interlayer coupling coefficient matrix and the interlayer contact area distribution map to obtain an interlayer interaction deformation model. Description of the Drawings

[0018] Figure 1 It is a schematic diagram of the step flow of a shoe upper contour detection method based on an image recognition model;

[0019] Figure 2 It is a detailed implementation step flow schematic diagram of step S1 in the present invention.

[0020] The realization of the purpose, functional characteristics and advantages of the present invention will be further described with reference to the embodiments and the accompanying drawings. Specific Embodiments

[0021] The technical method of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0022] In addition, the attached drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0023] It should be understood that although terms such as "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed related items.

[0024] In the embodiments of the present invention, reference Figure 1 As shown, it is a schematic diagram of the step flow of the shoe upper contour detection method based on an image recognition model of the present invention. In this example, the shoe upper contour detection method based on the image recognition model includes the following steps:

[0025] Step S1: Collect the original multi-state images of the shoe upper under different bending and extrusion degrees, and perform contour enhancement processing to obtain a set of deformed state contours;

[0026] In the embodiments of the present invention, first, a shoe upper deformation test platform is constructed. Through a biaxial rotation mechanism and a pneumatic compression device, the bending angle (0° - 45°, at intervals of 5°) and pressure (0 - 200 N, at intervals of 20 N) of the shoe upper are precisely controlled to form a 10x11 deformed state parameter matrix. Then, three high-definition cameras (side view, top view, 45° oblique view) are used to take pictures of the shoe upper fixed on this platform from different perspectives. Three images are collected under each parameter combination, and a total of 330 original images are obtained. These original images are preprocessed (brightness equalization, Gaussian blur filtering, geometric correction), the shoe upper area is segmented (adaptive threshold segmentation, morphological operations), and the fine contour line of the shoe upper is extracted (Canny edge detection, contour tracking, smoothing processing, digital encoding). Finally, the contour line data is associated with the deformed state parameter matrix to obtain a set of deformed state contours, and each contour line is labeled with a bending angle, pressure, and viewing angle.

[0027] Step S2: Identify and mark the key parts of the deformation state contour set to obtain a part marking map; accurately locate the part interface according to the part marking map to obtain an accurate interface map; construct a three-dimensional hierarchical structure map according to the accurate interface map;

[0028] In the embodiment of the present invention, the material texture features (LBP, HOG, GLCM, multi-scale) are extracted from the internal area of the contour in the deformation state contour set, and the mean shift clustering algorithm is used for the preliminary clustering of the material areas to obtain an initial material partition map. Combining the initial material partition map and the deformation state contour set, the candidate structural boundary points (curvature, gray gradient, non-maximum suppression, texture difference degree) are screened out. Based on the candidate structural boundary set and the initial material partition map, the part recognition and marking are carried out by using prior knowledge and shape descriptors (region connectivity analysis, shape descriptors, rule tables, manual verification) to obtain a part marking map. Then, the accurate positioning of the interface is carried out for the part marking map and the candidate structural boundary set (neighborhood search, orthogonal gradient field calculation, gradient peak tracking, sub-pixel edge accurate positioning, breakpoint detection and connection, local enhancement of fuzzy regions) to obtain an accurate interface map. Finally, according to the accurate interface map and the part marking map, the hierarchical superposition relationship is analyzed (displacement analysis, overlapping region detection, gradient direction analysis) to construct a three-dimensional hierarchical structure map, which contains information such as part boundary coordinates, hierarchical relationships, connection relationships, and connection strengths.

[0029] Step S3: Perform local key point deformation tracking according to the three-dimensional hierarchical structure map to obtain multi-dimensional deformation parameters; perform key part deformation analysis according to the multi-dimensional deformation parameter set to obtain a key point deformation characteristic table;

[0030] In the embodiment of the present invention, eight predefined key points (toe tip, heel, front edge of the tongue, root of the tongue, inner / outer side upper maximum width points, front / rear points of the sole bending area) are identified and located according to the three-dimensional hierarchical structure map, and their three-dimensional coordinates in the initial state are recorded. The local feature windows around each key point are extracted, and the feature vectors (HOG, neighborhood radius ratio, local curvature, corner response value, LBP) are calculated to obtain a key point feature description set. The deformation states are serially arranged (bending angle first, pressure second) to construct a deformation path sequence. Using an improved point-to-point registration algorithm, combining feature similarity and local deformation constraints, the key points in different deformation states are tracked to obtain a point tracking mapping table. The thin plate spline interpolation method is used to establish a non-rigid deformation field description, and the local affine transformation matrix and the relative motion consistency between key points are calculated. Finally, the multi-dimensional deformation parameters (displacement, curvature, strain, deformation rate) are calculated, and their variation laws with the bending angle and pressure are analyzed. The deformation analysis (bending characteristics, folding mode, stability evaluation) of the key parts (toe tip, tongue, heel) is carried out to obtain a key point deformation characteristic table.

[0031] Step S4: Identify the interlayer contact area distribution map based on the key point deformation characteristic table and the three-dimensional hierarchical structure map; analyze the interlayer stress transfer path of the interlayer contact area distribution map to obtain the stress transfer path map; calculate the interlayer coupling coefficient according to the stress transfer path map and the key point deformation characteristic table to obtain the interlayer coupling coefficient matrix; perform interlayer deformation response modeling according to the interlayer coupling coefficient matrix and the interlayer contact area distribution map to obtain the interlayer interaction deformation model;

[0032] In the embodiment of the present invention, based on the key point deformation characteristic table and the three-dimensional hierarchical structure map, the interlayer contact area (contact boundary, contact area, meshing, fixed connection points, sliding contact area, contact strength heat map) is identified to obtain the interlayer contact area distribution map. Calculate the strain gradient field (principal strain, secondary strain and their directions), and calculate the contact point stress in combination with the material constitutive relationship (Hooke's law, different Young's moduli are used for different materials) to obtain the contact point stress distribution table. Calculate the stress direction vector field, and extract the stress transfer path set accordingly. Identify the stress characteristic points (concentration points, dispersion points, turning points), and measure the stress delay time to obtain the stress delay coefficient table. Finally, construct the stress transfer path map to clearly show the path, direction, strength and delay of stress transfer. Then, perform the calculation of the interlayer coupling coefficient, including: stress path level division, deformation data pairing, calculation of deformation correlation coefficient, calculation of deformation phase difference, calculation of deformation transfer efficiency, and finally construct the interlayer coupling coefficient matrix. Finally, perform the interlayer deformation response modeling: establish the interlayer constraint equation (displacement continuity, friction model, deformation coordination, geometric compatibility, minimum principle of deformation energy), perform the analysis of deformation influence factors (control variable method, sensitivity coefficient, response time, critical deformation point, degree of restriction), extract the co-deformation mode (correlation coefficient, main / secondary mode, eigenvector, stability index, activation condition), construct the interlayer response function (piecewise linear interpolation, least squares method, cross-validation, regularization), and finally integrate to obtain the interlayer interaction deformation model to realize the prediction of the overall contour deformation of the shoe upper.

[0033] As an example of the present invention, refer to Figure 2 As shown, in this example, step S1 includes:

[0034] Step S11: Design the bending angle and pressure combination parameters of the shoe upper to obtain the deformation state parameter matrix;

[0035] In an embodiment of the present invention, a test platform for controlling the deformation of the shoe upper is constructed. The test platform includes a biaxial rotation mechanism for precisely controlling the bending angle of the shoe upper with an angle control accuracy of ±0.5°; and a pneumatic compression device for applying a uniform pressure with a pressure control accuracy of ±2N. Based on this platform, a two-dimensional parameter matrix is designed as the control parameter for the deformation of the shoe upper. The rows of the matrix represent the bending angles, ranging from 0° to 45°, at intervals of 5°, with a total of 10 rows. The columns of the matrix represent the pressures, ranging from 0N to 200N, at intervals of 20N, with a total of 11 columns. Therefore, the deformation state parameter matrix is a 10×11 matrix, containing a total of 110 elements, and each element represents a specific combination of bending angle and pressure, such as (0°, 0N), (0°, 20N)... (45°, 200N). This matrix will be used as the control basis for subsequent image acquisition.

[0036] Step S12: Acquire the original multi-state images of the shoe upper under different bending and extrusion degrees according to the deformation state parameter matrix;

[0037] In an embodiment of the present invention, the shoe upper sample to be tested is fixed on the test platform. According to the deformation state parameter matrix generated in step S11, the bending angle and pressure value of the platform are set sequentially. Under each parameter combination, the shoe upper images are taken from different perspectives using three high-definition cameras. The three cameras are respectively fixed on the side, top view, and 45° oblique view of the shoe upper to capture the deformation characteristics of the shoe upper in different directions. The pixel resolution of the cameras is set to 3840×2160. To ensure the image quality, a light source condition of 400lux uniform illumination is adopted. For each element in the deformation state parameter matrix (i.e., each combination of bending angle and pressure), three images (one for each of the three perspectives) are acquired. Therefore, after all the acquisitions are completed, an image library containing 330 original images (110 parameter combinations × 3 perspectives) will be obtained.

[0038] Step S13: Perform image preprocessing on the original multi-state images to obtain preprocessed images;

[0039] In an embodiment of the present invention, the 330 original images obtained in step S12 are preprocessed one by one. First, perform brightness equalization processing on each image, and use histogram stretching technology to stretch the pixel value range of the image to 0-255 to enhance the contrast of the image. Then, apply Gaussian blur filtering to remove the noise in the image, and the kernel size of the Gaussian blur is set to 5×5 pixels. Next, perform geometric correction, and through affine transformation, correct the images taken from different perspectives to a unified coordinate system to ensure the consistency of the position and posture of the shoe upper in different images. After the above processing, 330 preprocessed images are obtained.

[0040] Step S14: Segment the upper surface area of the preprocessed image to obtain the upper surface mask image;

[0041] In the embodiment of the present invention, the upper surface area of the preprocessed image obtained in step S13 is segmented. The adaptive threshold segmentation method is adopted to calculate an optimal segmentation threshold for each image. This threshold is automatically determined according to the local gray distribution of the image to adapt to the illumination changes of different images. According to the calculated threshold, the image pixels are divided into two categories: the pixels with pixel values greater than the threshold are classified as the upper surface area, and the pixels with pixel values less than the threshold are classified as the background area. Then, morphological operations are applied to optimize the segmentation result. First, an erosion operation is performed, and the size of the erosion kernel is 3×3 pixels to eliminate small noise points in the background. Then, a dilation operation is performed, and the size of the dilation kernel is also 3×3 pixels to fill small holes in the upper surface area. After the above processing, 330 upper surface mask images are obtained, and each image clearly marks the upper surface area (white) and the background area (black).

[0042] Step S15: Extract and refine the contour based on the upper surface mask image and the preprocessed image to obtain the fine contour line data;

[0043] In the embodiment of the present invention, based on the upper surface mask image generated in step S14 and the preprocessed image obtained in step S13, the fine contour line of the upper surface is extracted. First, the Canny edge detection algorithm is applied to the preprocessed image. The low threshold of the Canny algorithm is set to 50, and the high threshold is set to 150. This algorithm can detect the edge pixels with significant gray changes in the image. Then, the contour tracking algorithm is used to connect the detected edge pixels into continuous contour lines. For the broken contour lines, they are connected by calculating the distance and direction between the breakpoints. Next, the contour lines are smoothed by using the moving average method with a sliding window size of 5 points to reduce the jagged edges on the contour lines. Finally, each contour line is digitally encoded to record the coordinates (x, y) of the key inflection points on the contour line and the curvature information of each point. After the above processing, a fine contour line data set is obtained, which contains the data of 330 contour lines.

[0044] Step S16: Perform deformation state correlation integration on the fine contour line data and the deformation state parameter matrix to obtain the deformation state contour set;

[0045] In the embodiments of the present invention, the fine contour line dataset obtained in step S15 is associated with the deformation state parameter matrix generated in step S11. Two attribute tags, namely bending angle and pressure, are added to each contour line in the fine contour line dataset. The values of these two tags are determined according to the parameter combination used during the acquisition of the original image corresponding to the contour line. For example, if a certain contour line is extracted from an image acquired under the conditions of a bending angle of 10° and a pressure of 60 N, then the bending angle tag of this contour line is 10°, and the pressure tag is 60 N. In addition, a perspective tag is added according to the perspective (side view, top view, or 45° oblique view) to which the contour line belongs. A query index system is established, which can quickly retrieve the corresponding contour line data according to the input bending angle, pressure, and perspective. After the above processing, the final deformed state contour set is obtained. This contour set contains the upper surface contour information in all deformation states, and each contour line is associated with clear deformation parameters and perspectives.

[0046] Preferably, step S2 includes the following steps:

[0047] Step S21: Extract the material texture features from the deformed state contour set to obtain the upper surface texture feature map;

[0048] Step S22: Perform preliminary clustering of the material regions according to the upper surface texture feature map to obtain the initial material partition map;

[0049] Step S23: Screen the candidate structure boundary points according to the initial material partition map and the deformed state contour set to obtain the candidate structure boundary set;

[0050] Step S24: Perform part recognition and marking according to the candidate structure boundary set to obtain the part marking map;

[0051] Step S25: Perform precise positioning of the interface according to the candidate structure boundary set and the part marking map to obtain the precise interface map;

[0052] Step S26: Analyze the hierarchical superposition relationship according to the precise interface map and the part marking map to obtain the hierarchical relationship table;

[0053] Step S27: Construct a three-dimensional hierarchical structure map according to the hierarchical relationship table, the precise interface map, and the part marking map.

[0054] In the embodiments of the present invention, for the contour information included in the deformation state contour set obtained in step S1, the material texture features of the internal region of the contour are extracted. The Local Binary Pattern (LBP) algorithm is used to calculate the LBP feature value for each pixel point within the contour. The window size of the LBP algorithm is set to 8×8 pixels. For each central pixel, the gray values of its surrounding 8 pixels are compared with the gray value of the central pixel. Pixels with gray values greater than or equal to the central pixel's gray value are assigned 1, and those less than are assigned 0, thus obtaining an 8-bit binary code. This binary code is converted into a decimal number as the LBP feature value of the central pixel. In addition to the LBP feature, the Histogram of Oriented Gradients (HOG) feature is also calculated. The region within the contour is divided into 12 direction bins, and the gradient intensity in each direction is statistically analyzed. At the same time, four statistics of the Gray-Level Co-Occurrence Matrix (GLCM) are calculated: contrast, homogeneity, energy, and correlation. To capture texture information at different scales, texture features are extracted at three scales: 4×4 pixels, 8×8 pixels, and 16×16 pixels. The LBP feature, HOG feature, and GLCM feature are combined at the three scales to form a multi-scale texture descriptor. The multi-scale texture descriptor is calculated for all pixel points within the contour, and finally, the shoe upper texture feature map is obtained.

[0055] Based on the shoe upper texture feature map obtained in step S21, a preliminary material partition of the shoe upper region is performed. The MeanShift clustering algorithm is used. The bandwidth parameter of the MeanShift algorithm is set to 0.8. This algorithm iteratively calculates the feature mean of the region around each pixel point and moves the pixel point to the mean position until convergence. Through the MeanShift algorithm, pixel points with similar texture features are clustered together. The preset clustering number range is 3 - 5, and the optimal clustering number is automatically determined by calculating the intra-class similarity and inter-class difference. The intra-class similarity calculates the average distance of the feature vectors of pixel points within the same category, and the inter-class difference calculates the distance between the centers of different categories. The clustering result with the smallest intra-class similarity and the largest inter-class difference is selected. Each region obtained by clustering is marked with different labels to generate the initial material partition map.

[0056] Based on the initial material partition map obtained in step S22 and the deformation state contour set obtained in step S1, potential structural boundary points are screened. First, the boundary lines between different partitions in the initial material partition map are extracted as candidates for the structural boundary. The curvature of each point on the boundary line is calculated, and points with a curvature greater than 0.05 are marked as key feature points. These high-curvature points usually correspond to obvious turns or corners in the shoe upper structure. Then, the gray gradient intensity on the boundary line is calculated, and only boundary points with a gradient intensity greater than 60 are retained. This step can exclude weak boundaries caused by texture changes within the material. The non-maximum suppression algorithm is applied to search for local gradient maximum points in the normal direction of the boundary line with a search radius of 5 pixels, and only the points with the local maximum gradient are retained. To further verify the effectiveness of the boundary, the texture difference degree between the regions on both sides of the boundary line is calculated. The cosine similarity of the texture feature vectors (refer to S21) of the regions on both sides is calculated, and boundary points with a similarity less than 0.7 (i.e., a difference degree greater than 0.3) are retained. After the above screening, a set of structural boundary candidates is obtained.

[0057] Based on the set of structural boundary candidates obtained in step S23 and the initial material partition map obtained in step S22, different parts of the shoe upper are identified and marked. According to the prior knowledge of the shoe structure, the sole is usually located at the bottom of the image, and the tongue is usually located in the upper central area of the image. Using this position information, the regions enclosed by the set of structural boundary candidates are initially marked. Region connectivity analysis is applied to determine adjacent regions. Shape descriptors are used to assist in the identification. The slenderness ratio (ratio of the major axis to the minor axis), roundness (4π×area / perimeter²), and convexity (region area / convex hull area) of each region are calculated. For example, the tongue region usually has a relatively large slenderness ratio, and the sole region has a lower roundness. A part identification rule table is established, which contains a series of rules for determining the category (tongue, upper, or sole) to which each region belongs. The rules are based on the position of the region, shape descriptors, and adjacent relationships with other regions. According to the rule table, each region is marked. Finally, manual verification is performed to check and correct the results of the automatic identification to ensure the accuracy of the marking. After the above processing, a part marking map is obtained, in which each region is clearly marked as tongue, upper, or sole.

[0058] Based on the part marking map obtained in step S24 and the candidate set of structural boundaries obtained in step S23, accurately locate the interfaces between the tongue and the upper, and between the upper and the sole. First, extract the boundary segments between different parts in the part marking map, set the neighborhood search width to 5 pixels, and screen candidate structural boundary points within the label change area. Assign a unique identifier to each junction area. Then, establish a 5×5 pixel local calculation window around the junction area, apply Sobel operators in the horizontal and vertical directions to the image within the window to calculate the gradient components, synthesize the gradient vector field, and record the gradient intensity and direction of each pixel point. Calculate the orthogonal component of the gradient direction and the boundary tangent, and retain the points where the orthogonal component is greater than the gradient threshold of 30. Perform non-maximum suppression on the gradient field, with the suppression window size of 3×3 pixels, and retain the local maximum gradient points. Starting from one end of the junction area, track the gradient peak points along the gradient direction. Set the search angle range to ±30 degrees, and search for the next gradient peak point within this range. When tracking encounters a fork, select the direction with the maximum gradient intensity to continue. Record the coordinates of all peak points on the tracking path. Use the moving average method (window size of 5 points) to reduce path oscillation. For each point on the gradient peak chain, extract a one-dimensional gradient profile perpendicular to the edge direction (length 7 pixels). Use the parabolic fitting method to fit the gradient profile and calculate the offset of the fitting peak from the integer pixel point (accuracy 0.1 pixel). Update the edge point coordinates by adding the sub-pixel offset. Scan the boundary chain to detect the break point positions where the point spacing is greater than 2 pixels. For each pair of break points, calculate the direction vectors of the two end points. Use the cubic spline interpolation method to generate a connecting curve based on the position and direction information of 3 points on each side of the break point. Insert an appropriate number of intermediate points (spacing about 0.5 pixel) to ensure the continuity and smoothness of the boundary line. Identify the parts with a confidence level lower than 0.6 in the continuous boundary chain and mark them as fuzzy areas. For each fuzzy area, extract a local image patch of 11×11 pixels around it. Apply local contrast enhancement processing with a contrast enhancement coefficient of 1.5. Recalculate the gradient field of the enhanced area and update the boundary position and confidence level. Check whether the boundary line is consistent with the boundary of the part marking, and calculate the average deviation distance. Mark the areas with a deviation greater than 1 pixel. Use the local deformation algorithm to adjust the boundary line to a position more consistent with the part marking. Maintain the smoothness of the boundary line, calculate the local curvature, and limit the maximum curvature change rate. Apply the tension spring model (elastic coefficient 0.7) to all boundaries to balance the boundary accuracy and smoothness. Integrate the optimized boundary chains of all junction areas into one map, and assign different identifiers and visual representations to different types of interfaces (tongue-upper, upper-sole). Create a boundary attribute table to record information such as the type, length, and average confidence level of each boundary segment. Generate the boundary network topology to record the connection relationships of the interfaces. Build a multi-level representation to support the viewing of boundaries at different precision levels, and obtain the accurate interface map.

[0059] Based on the accurate interface diagram obtained in step S25 and the part marking diagram obtained in step S24, analyze the hierarchical superposition relationship between different parts of the shoe upper. Analyze the displacement of the interface under different deformation states (refer to step S1), compare the movement trajectories of different parts, and judge their relative movement relationship. For example, if the tongue moves upward relative to the vamp, it indicates that the tongue is above the vamp. Detect the overlapping area near the interface and observe the brightness change and shadow characteristics of the overlapping area. Usually, the edge of the upper structure will block the lower structure, resulting in a darker area or shadow of the lower structure in the image. Based on these brightness changes and shadows, judge the upper and lower relationship of the layers. Calculate the gradient direction of the upper edge points on the interface. If the gradient direction points from the inside to the outside, the edge is a convex edge, indicating that this part is on the upper layer; if the gradient direction points from the outside to the inside, the edge is a concave edge, indicating that this part is on the lower layer. Based on the above analysis, establish a hierarchical relationship table. This table records the hierarchical relationship between each part of the shoe upper (tongue, vamp, sole). For each pair of adjacent parts (such as tongue-vamp, vamp-sole), record their upper and lower layer relationship (for example, the tongue is above the vamp). At the same time, record the relative depth value of each part, set the depth value of the outermost part to 0, and increase it sequentially inward. For example, if the tongue is above the vamp and the vamp is above the sole, the depth value of the tongue is 0, the depth value of the vamp is 1, and the depth value of the sole is 2. The hierarchical relationship table stores this information in the form of a table.

[0060] Based on the hierarchical relationship table obtained in step S26, the precise interface diagram obtained in step S25, and the part marking diagram obtained in step S24, construct a three-dimensional hierarchical structure atlas that can describe the spatial relationships of each part of the shoe upper. First, according to the precise interface diagram, extract the boundary coordinates of each part. These coordinates are two-dimensional (x, y), representing the positions on the image plane. Then, combine the depth information in the hierarchical relationship table to add a depth coordinate (z) to the boundary points of each part. The value of the z coordinate is determined by the relative depth value of the part. For example, the z coordinate of the tongue is 0, the z coordinate of the vamp is 1, and the z coordinate of the sole is 2. In this way, the two-dimensional boundary coordinates are extended to three-dimensional coordinates (x, y, z). Based on these three-dimensional coordinates, establish the spatial topological relationships between parts. For adjacent parts, determine the connection relationships between them. The connection relationships can be rigid (such as suture connection), flexible (such as the natural curvature of leather), or semi-rigid (such as adhesive connection). Calculate the relative displacement vectors of each part in different deformation states. These vectors describe the movement directions and amplitudes of each part during the deformation process. Draw a part connection diagram in three-dimensional space. The nodes in the diagram represent different parts of the shoe upper (tongue, vamp, sole), and the positions of the nodes are determined by their three-dimensional coordinates. The lines between the nodes represent the connection relationships between parts, the directions of the lines represent the connection directions, and the thicknesses of the lines represent the connection strengths. The connection strength is determined by calculating the correlation of the displacements at the connection points in different deformation states. The higher the correlation, the greater the connection strength, ranging from 0 to 1. The connection methods are represented by different line types (solid lines represent rigid connections, dashed lines represent flexible connections, and dotted lines represent semi-rigid connections). Integrate the spatial relationship data in different deformation states to generate a complete three-dimensional hierarchical structure atlas. This atlas contains information such as the boundary coordinates, hierarchical relationships, connection relationships, and connection strengths of each part of the shoe upper, and can comprehensively describe the spatial composition of the shoe upper structure under different deformation conditions.

[0061] Preferably, step S25 includes the following steps:

[0062] Step S251: Extract the junction area location map according to the structural boundary candidate set and the part marking diagram;

[0063] Step S252: Perform orthogonal gradient field calculation on the junction area location map to obtain an enhanced gradient field map;

[0064] Step S253: Perform gradient peak tracking on the enhanced gradient field map to obtain a gradient peak chain;

[0065] Step S254: Perform sub-pixel edge precise positioning on the enhanced gradient field map according to the gradient peak chain to obtain a sub-pixel precise boundary;

[0066] Step S255: Perform break point detection and connection on the sub-pixel precise boundary to obtain a continuous boundary chain;

[0067] Step S256: Perform local enhancement of the fuzzy region based on the continuous boundary chain and the part marking map to obtain an accurate interface map.

[0068] In the embodiment of the present invention, based on the structural boundary candidate set obtained in step S23 (including potential boundary points and their coordinates) and the part marking map obtained in step S24 (defining regions such as the tongue, upper, and sole and their labels), the initial positioning of the junction region is extracted. Traverse the part marking map to find positions where adjacent pixel points have different labels. These positions represent the junctions between different parts. For example, if a pixel point is marked as "tongue" and its adjacent pixel point is marked as "upper", then the boundary between these two pixel points is the junction between the tongue and the upper. For each junction, expand 5 pixels on both sides to form a strip-shaped region as the junction region. Extract the boundary points in the structural boundary candidate set that are located within this junction region as the candidate boundary points for this junction region. Assign a unique identifier to each junction region (such as the tongue-upper junction region, the upper-sole junction region), and record its starting point and ending point coordinates, the set of candidate boundary points, and the pair of parts to which it belongs. Integrate this information to generate a junction region positioning map. This map contains the preliminary position and range information of all junction regions.

[0069] Based on the junction region positioning map obtained in step S251, calculate the orthogonal gradient field for the image within each junction region. First, crop the part of the original image (the preprocessed image from step S13) corresponding to the junction region. Then, calculate the gradient of the cropped image block. Use the Sobel operator to calculate the gradient components of the image in the horizontal and vertical directions. The Sobel operator in the horizontal direction is [-1, 0, 1; -2, 0, 2; -1, 0, 1], and the Sobel operator in the vertical direction is [-1, -2, -1; 0, 0, 0; 1, 2, 1]. Convolve the image with these two operators respectively to obtain the horizontal gradient Gx and the vertical gradient Gy. For each pixel point, calculate its gradient intensity: G = √(Gx 2+Gy2), and the gradient direction: θ = arctan(Gy / Gx). Then, calculate the angle between the gradient direction θ and the tangent direction of the initial boundary line of the junction region at this point. The initial boundary line is formed by connecting the candidate boundary points extracted in step S251. Calculate the sine value sin(α) of this angle, where α is the angle between the gradient direction and the tangent direction. Multiply the gradient intensity G by sin(α) to obtain the orthogonal gradient component. Only retain the pixel points with the orthogonal gradient component greater than 30. Finally, perform non-maximum suppression on the retained pixel points. In the gradient direction of each pixel point, compare its orthogonal gradient component with the adjacent two pixel points, and only retain the local maximum points. After the above processing, an enhanced gradient field map is obtained. This map highlights the gradient information perpendicular to the boundary of the junction region.

[0070] Based on the enhanced gradient field map obtained in step S252, perform gradient peak tracking within each junction region to obtain a preliminary boundary line. Start from one end of the junction region (the starting point is determined by step S251), and search for the gradient peak points along the gradient direction. At each current point, calculate its gradient direction. Set a search angle range of ±30 degrees centered on this direction. Within this range, find the next pixel point with the maximum gradient intensity as the new current point. Add the new current point to the tracking path. Repeat the above process until reaching the other end of the junction region (the termination point is determined by step S251) or there are no gradient peak points within the search range. If a gradient bifurcation is encountered (i.e., there are multiple points with similar gradient intensities), select the direction with the maximum gradient intensity to continue tracking. Record the coordinates of all pixel points on the tracking path to form a preliminary boundary line, that is, the gradient peak chain. Smooth the obtained gradient peak chain. Use the moving average method to average the coordinates of each point, and the window size is 5 points.

[0071] Based on the gradient peak chain obtained in step S253 and the enhanced gradient field map obtained in step S252, perform sub-pixel level precise positioning of the boundary points. For each point on the gradient peak chain, extract a one-dimensional gray level profile with a length of 7 pixels in the direction perpendicular to its gradient direction (calculated in step S252). This profile takes the current point as the center and takes 3 pixels on each side. Fit a parabola to this one-dimensional gray level profile. Assume the parabola equation is y = ax 2+bx + c, where y represents the grayscale value and x represents the pixel position. Using the least squares method, the coefficients a, b, and c of the parabolic equation are solved. Calculate the vertex coordinates of the parabola x_vertex = -b / (2a). The x coordinate of this vertex is the edge position at the sub-pixel level. Calculate the difference between x_vertex and the integer pixel coordinate of the current point, and this difference is the sub-pixel offset. Add this offset to the integer pixel coordinate of the current point to obtain the accurate coordinate at the sub-pixel level. Perform the above processing on all points on the gradient peak chain to obtain the sub-pixel accurate boundary.

[0072] Based on the sub-pixel accurate boundary obtained in step S254, detect and connect the breakpoints in the boundary. Traverse all points on the sub-pixel accurate boundary and calculate the distance between adjacent points. If the distance is greater than 2 pixels, it is considered that there is a breakpoint between these two points. For each pair of detected breakpoints, calculate the average direction vectors of 3 points on each side of the breakpoint. Use the cubic spline interpolation method to generate a smooth curve connecting the breakpoints according to these two direction vectors and the position of the breakpoint. On this curve, insert new points at intervals of 0.5 pixels to fill the gap between the breakpoints. Add the newly inserted points to the sub-pixel accurate boundary to form a continuous boundary chain.

[0073] Based on the continuous boundary chain obtained in step S255 and the part marking map obtained in step S24, perform local enhancement on the blurred area on the boundary. First, calculate the confidence of each point on the continuous boundary chain. The confidence is determined based on the goodness of fit (R 2 value) of the parabolic fitting in step S254. The closer the R 2 value is to 1, the better the fitting effect and the higher the confidence. Mark the points with a confidence lower than 0.6 as blurred points. For each blurred point, extract the local image block of 11×11 pixels around it (from the preprocessed image in step S13). Perform local contrast enhancement on this image block. Use the CLAHE (Contrast Limited Adaptive Histogram Equalization) algorithm. Divide the image block into 8×8 sub-blocks, calculate the histogram for each sub-block, and perform histogram equalization. Limit the contrast enhancement amplitude of each sub-block to avoid over-enhancing noise. Stitch the enhanced sub-blocks together to obtain the enhanced local image block. For the enhanced image block, repeat steps S252, S253, and S254 to recalculate the gradient field, gradient peak chain, and sub-pixel accurate boundary of the blurred area. Replace the original blurred points with the newly calculated boundary points. Merge all the updated boundary points and the un-enhanced boundary points to obtain the final accurate interface map.

[0074] Preferably, the local key point deformation tracking described in step S3 includes:

[0075] Perform key point recognition and positioning according to the three-dimensional hierarchical atlas to obtain a key point coordinate library;

[0076] Extract local feature windows based on the key point coordinate library and the deformation state contour set to obtain a key point feature description set;

[0077] Perform deformation serialization arrangement according to the key point coordinate library and the deformation state contour set to obtain a deformation path sequence;

[0078] Perform point-to-point registration tracking according to the deformation path sequence and the key point feature description set to obtain a point tracking mapping table;

[0079] Perform non-rigid deformation analysis according to the point tracking mapping table and the deformation path sequence to obtain a non-rigid deformation field description;

[0080] Perform deformation parameter quantization calculation according to the non-rigid deformation field description and the key point coordinate library to obtain multi-dimensional deformation parameters.

[0081] In the embodiment of the present invention, based on the three-dimensional hierarchical atlas, key points on the shoe upper are recognized and positioned. According to the structural characteristics of the shoe, 8 key points are predefined: the shoe tip (the point with the largest front-end curvature), the shoe heel (the center of curvature of the rear end), the front edge of the tongue (the center point of the free edge in the middle and upper part), the root of the tongue (the center point at the junction of the tongue and the shoe upper), the point with the maximum width of the inner shoe upper, the point with the maximum width of the outer shoe upper, the front point of the sole bending area (the point with the largest bottom curvature change), and the rear point of the sole bending area. For the shoe tip and the shoe heel, calculate the curvatures of the front-end and rear-end contour lines of the shoe upper in the three-dimensional hierarchical atlas. The curvature is obtained by calculating the reciprocal of the radius of the circular arc formed by each point on the contour line and several adjacent points. Find the point with the largest curvature as the shoe tip, and find the center point of curvature at the rear end as the shoe heel. For the front edge of the tongue and the root of the tongue, determine the tongue area according to the part marking map (step S24). The midpoint of the upper edge in the tongue area is used as the front edge of the tongue, and the midpoint of the junction line between the tongue and the shoe upper (step S25) is used as the root of the tongue. For the points with the maximum width of the inner and outer shoe uppers, calculate the width of the shoe upper area in the horizontal direction, and find the positions with the maximum width as the points with the maximum width of the inner and outer shoe uppers respectively. For the sole bending area, calculate the curvature change rate of the sole contour line, and find the two points with the largest change rate as the front and rear points of the sole bending area. Assign a unique identifier (such as ID1-ID8) to each key point. Record the three-dimensional coordinates (x, y, z) of these key points in the three-dimensional hierarchical atlas in the initial state (0° bending, 0N pressure) to generate a key point coordinate library.

[0082] Based on the key-point coordinate library and the set of deformed state contours, local features around each key point are extracted. For each key point, a circular region with a radius of 10 pixels is extracted on the image where it is located, centered on the key point, as the feature window of the key point. Feature extraction is performed on the image within the feature window. The histogram of oriented gradients (HOG) features within the feature window is calculated. The feature window is divided into gradient direction bins in 8 directions, and the gradient intensity in each direction is statistically calculated. A 50-dimensional feature vector is constructed. This vector includes: HOG features (8 dimensions), the ratio of the neighborhood radius (the ratio of the average value to the maximum value of the distances from the key point to all pixel points within its 10-pixel neighborhood, 1 dimension), local curvature (1 dimension), corner response value (calculated by the Harris corner detection algorithm, 1 dimension), and other texture features (such as local binary pattern LBP features, the LBP feature histogram within the feature window is statistically calculated, 39 dimensions). The feature vector is normalized, and the values in each feature dimension are scaled to between 0 and 1. The above operations are repeated for all key points and images in all deformation states, and a key-point feature description set is obtained. This set contains the feature vectors of each key point in each deformation state.

[0083] Based on the set of deformed state contours and the key-point coordinate library obtained in step S1, all deformation states are sorted to construct a continuous deformation sequence. All deformation states are sorted according to the bending angle and the pressure value. First, they are sorted in ascending order of the bending angle (0° - 45°, with an interval of 5°), and then, at each bending angle, they are sorted in ascending order of the pressure value (0 - 200 N, with an interval of 20 N). In this way, a deformation state sequence containing 110 elements is obtained. A deformation state index table is constructed, which records the position of each deformation state (uniquely determined by the bending angle and the pressure value) in the sequence. The average contour difference value between two adjacent deformation states in the sequence is calculated. This difference value is obtained by calculating the average value of the Euclidean distances between all corresponding contour points in the two states. If the difference value between two adjacent states exceeds 0.15, it is considered that there is a deformation step point between these two states. A deformation process path diagram is generated. In this diagram, the bending angle and the pressure value are used as the coordinate axes, and all deformation states are represented as nodes in the diagram. According to the order of the deformation sequence, adjacent nodes are connected by directed line segments to form a path from the initial state (0° bending, 0 N pressure) to the final state (45° bending, 200 N pressure). This path represents the optimal transition sequence of the upper deformation.

[0084] Based on the deformation path sequence and the key point feature description set, track the key points in different deformation states. An improved point-to-point registration algorithm is adopted. For two adjacent deformation states in the deformation path sequence, calculate the feature similarity between all key points in these two states. The feature similarity is obtained by calculating the cosine similarity between the feature vectors of two key points. Only retain the key point pairs with a similarity greater than 0.75. For each key point, in the image of its next deformation state, search for the point with the highest feature similarity within a range of 5 pixels around its predicted position (predicted based on the motion trend of the previous several states) as the corresponding point of this key point. Introduce a local deformation constraint matrix. This matrix describes the relative position relationship between each key point and the key points around it. During the registration process, limit the displacement of the key points to satisfy the local deformation constraints. For the key points that are occluded or disappeared in a certain deformation state, adopt a temporary point prediction strategy. According to the motion trajectory of this key point in the previous several states, predict its position in this state. If no point with similar features is found near the predicted position, use this predicted position as the temporary position of this key point. Repeat the above operations for all adjacent deformation states to obtain a complete point tracking mapping table. This table records the coordinates of the corresponding points (or temporary points) of each key point in all deformation states.

[0085] Based on the point tracking mapping table and the deformation path sequence, establish a mathematical model for the non-rigid deformation of the shoe upper. The Thin Plate Spline (TPS) method is adopted. TPS is an interpolation method based on radial basis functions, which can generate a smooth deformation field according to the displacements of a set of control points (key points). Use the key point coordinate library in each deformation state as the control points, and use the displacements of the key points between adjacent deformation states as the displacements of the control points. Use TPS interpolation to calculate the deformation field of the entire image area (not just the key points). This deformation field describes the displacement of each pixel point during the deformation process. Calculate the local affine transformation matrix within the neighborhood (radius 10 pixels) of each key point. This matrix describes the local deformation within this neighborhood, including rotation, scaling, and shear. Decompose the affine transformation matrix into a rotation matrix and a stretching matrix to separate the rigid deformation (rotation) and the non-rigid deformation (stretching and shear). Calculate the relative motion consistency between key points. For each pair of key points, calculate the correlation coefficient of their displacement vectors in all deformation states. The higher the correlation coefficient, the higher the motion consistency of these two key points. Based on the correlation coefficient, identify the deformation coordination regions of the shoe upper. Establish a coherent model for the deformation of each part of the shoe upper. This model contains the deformation field, the local affine transformation matrix, and the relative motion consistency information between key points in all deformation states.

[0086] Based on the non-rigid deformation field description and the key point coordinate library, various deformation parameters of the shoe upper during the deformation process are calculated. For each key point, its three-dimensional displacement vector (components in the x, y, and z directions) in each deformation state is calculated. The displacement vector is obtained by calculating the difference between the coordinates of the key point in that state and the coordinates in the initial state. Measure the local curvature change. For each key point, within its neighborhood (radius 10 pixels), fit a circular arc. Calculate the radius of the circular arc and take its reciprocal as the curvature of the point. Calculate the difference between the curvature of the key point in each deformation state and the curvature in the initial state. Estimate the surface strain distribution. Based on the key points, construct a triangular mesh. Calculate the area change of each triangle during the deformation process. The area change rate is used as the surface strain of the triangle. Calculate the principal strain (maximum strain) and the secondary strain (minimum strain). Determine the deformation rate. For each key point, calculate its displacement under unit bending angle change and unit pressure change. Calculate the variation laws of various deformation parameters (displacement, curvature, strain) with the bending angle and pressure. Use quadratic or cubic polynomial functions to fit these variation laws. Integrate all the calculated deformation parameters to obtain a multi-dimensional deformation parameter set. This set contains information such as the displacement, curvature, strain, and deformation rate of each key point in each deformation state.

[0087] Preferably, the key part deformation analysis described in step S3 includes:

[0088] Extract key part parameters according to the multi-dimensional deformation parameter set and the non-rigid deformation field description, where the key parts include toe parameters, tongue parameters, and heel parameters;

[0089] Calculate the bending characteristics of the toe parameters to obtain toe bending data; perform dynamic tracking of the buckling point based on the toe bending data and the toe parameters to obtain the dynamic characteristics table of the buckling point;

[0090] Perform tongue folding pattern recognition on the tongue parameters to obtain tongue fold line distribution data; measure the folding depth and angle based on the tongue fold line distribution data to obtain the tongue folding characteristics table;

[0091] Perform heel stability assessment on the heel parameters to obtain the heel stability index table; calculate the shape retention rate based on the heel stability index table to obtain the shape retention rate distribution data;

[0092] Perform inter-part deformation correlation analysis based on the dynamic characteristics table of the buckling point, the tongue folding characteristics table, and the shape retention rate distribution data to obtain the deformation transfer chain data;

[0093] Generate the key point deformation characteristics table based on the deformation transfer chain data and the multi-dimensional deformation parameters.

[0094] In the embodiments of the present invention, based on the multi-dimensional deformation parameter set and the non-rigid deformation field description, the deformation parameters related to three key parts, namely the shoe tip, the tongue and the heel, are extracted. According to the definition and identifier of the key points, the data of the key points belonging to these three parts are screened out from the multi-dimensional deformation parameter set. The shoe tip parameters include: the displacement, curvature and strain data of the key point of the shoe tip (ID1). The tongue parameters include: the displacement and curvature data of the front edge key point (ID3) and the root key point (ID4) of the tongue, and the non-rigid deformation field data of the tongue area (determined according to the part marking map). The heel parameters include: the displacement, curvature and strain data of the key point of the heel (ID2). These parameters are stored separately to form parameter sets for the three parts.

[0095] Analyze the extracted shoe tip parameters to evaluate the performance of the shoe tip during the bending process. Extract the displacement data of the key points of the shoe tip at different bending angles (0° - 45°, with an interval of 5°). Calculate the Pearson correlation coefficient between the displacement of the shoe tip (displacement relative to the initial position) and the bending angle. The value range of the correlation coefficient is from -1 to 1, and the closer it is to 1 or -1, the stronger the correlation. Identify the buckling point of the shoe upper. The buckling point is defined as the point with the maximum curvature on the front contour line of the shoe upper. At each bending angle, calculate the curvature of each point on the contour line of the shoe tip area (determined according to the fine contour line data), and find the point with the maximum curvature as the buckling point. Record the position (x, y coordinates) of the buckling point and its change with the bending angle. Measure the forward movement distance of the buckling point relative to the initial position, and calculate the proportional relationship between this distance and the bending angle. Analyze the linear / non-linear characteristics of the shoe tip deformation. According to the shape of the shoe tip displacement - bending angle curve, judge whether the deformation is linear or non-linear. Determine the critical bending angle. The critical bending angle is defined as the angle value at which the deformation characteristics change significantly, such as the angle at which the deformation changes from linear to non-linear. Integrate the above analysis results (correlation coefficient, buckling point position, forward movement distance, linear / non-linear characteristics, critical bending angle) to generate the shoe tip bending data.

[0096] Based on the toe bend data and toe parameters, track the dynamic behavior of the buckling point throughout the deformation process. Establish a correspondence table between the buckling point position (x, y coordinates) and the bending angle. Calculate the movement rate of the buckling point position with respect to the change in bending angle (the displacement amount caused by a unit change in angle). Analyze the smoothness of the buckling point trajectory. Calculate the continuity index of the buckling point trajectory curve. For example, calculate the standard deviation of the distances between adjacent points on the trajectory. The smaller the standard deviation, the smoother the trajectory. Measure the strain concentration degree in the area around the buckling point. At each bending angle, extract a circular area with a radius of 5 mm around the buckling point and calculate the average strain and maximum strain within this area. Identify the high-strain areas (areas where the strain value exceeds twice the standard deviation of the average strain). Evaluate the influence of different pressure values (0 - 200 N, at intervals of 20 N) on the buckling point position. At each pressure value, repeat the above analysis to construct a pressure-position influence matrix. The rows of this matrix represent the pressure values, the columns represent the bending angles, and the elements represent the positions of the buckling point. Integrate the above analysis results (movement rate, trajectory smoothness, strain concentration degree, pressure-position influence matrix) to generate a buckling point dynamic characteristic table.

[0097] Analyze the extracted tongue parameters to identify the folding pattern of the tongue during bending. According to the non-rigid deformation field description, calculate the curvature change map of the tongue area at each deformation state. The curvature change is obtained by calculating the difference in the relative displacement vectors between each pixel point and its surrounding pixel points. Identify the lines with a curvature value greater than 0.05 as candidate fold lines. According to the curvature magnitude, divide the fold lines into two categories: main fold lines (curvature greater than 0.08) and secondary fold lines (curvature between 0.05 - 0.08). Measure the length, direction (angle with the horizontal direction), and depth (the maximum distance from the fold line to the original tongue surface) of each fold line. Calculate the angular relationship between the fold lines. Identify parallel fold line groups (fold lines with an angle difference less than 10 degrees) and intersecting fold line groups (fold lines with an angle difference greater than 30 degrees). Analyze the timing of the appearance of the fold lines. Record at which bending angle and pressure conditions each fold line first appears. Determine the conditions for the formation of the fold lines (thresholds of bending angle and pressure). Integrate the above analysis results (fold line type, length, direction, depth, angular relationship, appearance timing, formation conditions) to generate tongue fold line distribution data.

[0098] Based on the distribution data of the tongue fold lines, accurately measure the folding depth and angle of the tongue. For each main fold line, measure its maximum folding depth. The maximum folding depth is defined as the maximum vertical distance from a point on the fold line to the original tongue surface (the surface in the undeformed state). Calculate the folding angle. The folding angle is defined as the included angle between the tongue surfaces on both sides of the fold line at this point. Calculate the cosine value of the included angle by taking the dot product of the normal vectors of the surfaces on both sides of the fold line, and then calculate the arccosine value to obtain the included angle. Analyze the variation laws of the folding depth and angle with the bending angle and pressure value. Establish the corresponding curves of folding depth - bending angle and folding angle - bending angle. Fit these curves using a quadratic polynomial function. Calculate the surface area change rate of the folding area. The surface area change rate reflects the degree of local material accumulation. Analyze the elastic recovery characteristics of the tongue material. After the pressure decreases, observe the persistence of the creases. Measure the residual depth and angle of the creases. Integrate the above analysis results (maximum folding depth, folding angle, depth - angle curve, surface area change rate, elastic recovery characteristics) to generate a table of tongue folding characteristics.

[0099] Analyze the extracted heel parameters to evaluate the stability of the heel under loading conditions. Measure the total displacement of the key points of the heel under different pressure and bending angle conditions. Calculate the maximum displacement vector of the key points of the heel relative to the initial position. Analyze the shape change of the heel contour. Based on the heel contour in the initial state, calculate the average offset distance of the corresponding points on the heel contour in each deformed state (determined according to the fine contour line data). Define the shape retention rate. Shape retention rate = (1 - average offset distance / heel characteristic dimension). Here, the heel characteristic dimension can be the height of the heel. The value range of the shape retention rate is 0 - 1, and the closer it is to 1, the better the shape is retained. Measure the stiffness coefficients of the heel in the transverse (horizontal direction) and longitudinal (vertical direction). The stiffness coefficient is defined as the displacement caused by a unit force (pressure or bending moment). Construct a heel stiffness distribution map. This map uses the transverse stiffness and longitudinal stiffness as coordinate axes to show the stiffness distribution of the heel in different deformed states. Integrate the above analysis results (total displacement, maximum displacement vector, average offset distance, shape retention rate, stiffness coefficient, stiffness distribution map) to generate a table of heel stability indicators.

[0100] Based on the heel stability index table, the shape retention rate of the heel is calculated in detail. The heel contour (according to the fine contour line data) in the initial state (0° bending, 0N pressure) is discretized into 50 uniformly distributed points. Record the coordinates of these points in the initial state. For each deformation state (uniquely determined by the bending angle and pressure value), according to the description of the non-rigid deformation field, calculate the new coordinates of these 50 points in this state. Calculate the Euclidean distance between the new coordinates and the initial coordinates of each point. Calculate the average displacement ratio of all 50 points. Average displacement ratio = sum of the displacement distances of all points / (50 × heel length). Where the heel length is defined as the length of the heel contour line in the initial state. Define the shape retention rate = 1 - average displacement ratio. The value range of the shape retention rate is 0 - 1. 1 means completely maintaining the original state, and 0 means completely deformed. Analyze the variation trend of the shape retention rate with pressure and bending angle. Plot the shape retention rate - pressure curve and the shape retention rate - bending angle curve. Determine the critical instability point. The critical instability point is defined as the point where the shape retention rate starts to decrease significantly (for example, the shape retention rate is lower than 0.9). Integrate the above analysis results (discrete point coordinates, displacement distances, average displacement ratio, shape retention rate, variation trend, critical instability point) to generate the shape retention rate distribution data.

[0101] Based on the dynamic characteristic table of the buckling point, the folding characteristic table of the tongue, and the shape retention rate distribution data, analyze the mutual influence and collaborative relationship among the three key parts, namely the toe, the tongue, and the heel, during the deformation process. Calculate the correlation matrix among the deformation parameters of the three key parts. Select representative parameters, such as: the buckling point position of the toe, the maximum folding depth of the tongue, and the shape retention rate of the heel. Calculate the correlation between these parameters pairwise using the Pearson correlation coefficient. The value range of the correlation coefficient is from -1 to 1, and the larger the absolute value, the stronger the correlation. Identify strongly correlated parameter pairs (parameter pairs with an absolute value of the correlation coefficient greater than 0.7). For example, if the correlation coefficient between the buckling point position of the toe and the maximum folding depth of the tongue is -0.8, it indicates a strong negative correlation between these two parameters, that is, the more obvious the buckling of the toe, the deeper the folding of the tongue. Determine the deformation transfer order. Analyze which part deforms before or after other parts. Judge by comparing the time series (bending angle and pressure value) at which the deformation parameters of different parts reach their maximum or minimum values. For example, if the buckling point position of the toe reaches its maximum value when the bending angle is 10°, and the maximum folding depth of the tongue reaches its maximum value when the bending angle is 15°, it indicates that the buckling of the toe precedes the folding of the tongue. Measure the time delay of the deformation transfer. The time delay is defined as the parameter increment (the difference in bending angle or pressure value) between the changes in the deformation parameters of two related parts. Construct a schematic diagram of the deformation transfer chain. This diagram uses the toe, the tongue, and the heel as nodes, and uses arrows to represent the direction of deformation transfer. The thickness of the arrow represents the influence intensity (the absolute value of the correlation coefficient), and the color of the arrow represents the positive or negative nature of the influence (red for positive correlation and blue for negative correlation). Calculate the contribution ratio of each part in the overall deformation. Determine by analyzing the explained variance of the deformation parameters of each part to the overall deformation (such as the change in the overall contour of the upper). Integrate the above analysis results (correlation matrix, strongly correlated parameter pairs, deformation transfer order, time delay, schematic diagram of the deformation transfer chain, contribution ratio) to generate deformation transfer chain data.

[0102] Generate a comprehensive key-point deformation characteristic table based on the deformation transfer chain data and multi-dimensional deformation parameters. This table summarizes the deformation characteristics of all key points under different deformation conditions. The rows of the table represent different deformation states (uniquely determined by the bending angle and pressure value), and the columns represent different key points and their deformation parameters. The key points include: the toe tip (ID1), the heel (ID2), the front edge of the tongue (ID3), the root of the tongue (ID4), the point of maximum width on the inner side of the shoe upper (ID5), the point of maximum width on the outer side of the shoe upper (ID6), the front point of the sole bending area (ID7), and the rear point of the sole bending area (ID8). The deformation parameters include: displacement (x, y, and z direction components), curvature, strain (principal strain, secondary strain), and deformation rate (the amount of displacement caused by a unit change in the bending angle, the amount of displacement caused by a unit change in pressure). In the table, fill in the corresponding deformation parameter values of each key point in each deformation state. These values are from the multi-dimensional deformation parameter set. Add the deformation correlation information between parts. Add columns to the table to record the correlation coefficient, deformation transfer order, and time delay between each key point and other key points. This information is from the deformation transfer chain data. Generate the key-point deformation characteristic table.

[0103] Preferably, the analysis of the interlayer stress transfer path in step S4 includes:

[0104] Identify the interlayer contact area distribution map according to the key-point deformation characteristic table and the three-dimensional hierarchical structure map;

[0105] Calculate the strain gradient according to the key-point deformation characteristic table to obtain the strain gradient field distribution map;

[0106] Calculate the contact point stress according to the strain gradient field distribution map and the interlayer contact area distribution map to obtain the contact point stress distribution table;

[0107] Calculate the stress direction vector field according to the contact point stress distribution table;

[0108] Extract the stress transfer path set of the contact point stress distribution table according to the stress direction vector field;

[0109] Identify the stress characteristic points according to the stress transfer path set and the stress direction vector field to obtain the stress characteristic point distribution map;

[0110] Measure the stress delay according to the stress characteristic point distribution map to obtain the stress delay coefficient table;

[0111] Construct the stress transfer path map according to the stress delay coefficient table, the stress characteristic point distribution map, and the stress transfer path set.

[0112] In the embodiments of the present invention, based on the key point deformation characteristic table and the three-dimensional hierarchical structure map, the contact areas between different layers of the shoe upper are identified. According to the hierarchical relationship and boundary coordinates of each part (tongue, vamp, sole) recorded in the three-dimensional hierarchical structure map, the contact surfaces between adjacent layers are determined. For example, there are contact surfaces between the tongue and the vamp, and between the vamp and the sole. For each pair of contact surfaces, the length of the contact boundary and the area of the contact area are calculated. The contact boundary is obtained by extracting the overlapping part of the boundaries of the two parts. The contact area is obtained by calculating the overlapping area of the projections of the two parts in the three-dimensional space. The contact area is gridded, and the contact area is divided into grid cells of 2mm×2mm. The fixed connection points and the sliding contact areas are marked. The fixed connection points refer to the points connected by sutures or adhesives, and these points will not undergo relative displacement during the deformation process. The sliding contact area refers to the area that is not fixedly connected but may come into contact and slide during the deformation process. The contact intensity heat map is drawn. According to the displacement data of each part recorded in the key point deformation characteristic table in different deformation states, the relative displacement of each point in the contact area is calculated. The smaller the relative displacement, the closer the contact and the greater the contact intensity. The contact intensity is represented by the shade of color, ranging from 0 to 1, where 0 represents complete separation and 1 represents complete contact. Integrate the above information (contact boundary, contact area, grid cells, fixed connection points, sliding contact areas, contact intensity heat map) to generate the interlayer contact area distribution map.

[0113] Based on the key point deformation characteristic table, the strain gradient of the shoe upper in different deformation states is calculated. For each key point, all pixel points (according to the image data) are extracted within its neighborhood (a circular area with a radius of 5mm). According to the displacement data of these pixel points recorded in the key point deformation characteristic table in different deformation states, the strain of each pixel point is calculated. The strain is obtained by calculating the relative displacement between this pixel point and its surrounding pixel points. The strain tensor includes normal strain (tensile or compressive) and shear strain (shearing strain). The principal strain (maximum strain) and secondary strain (minimum strain) of each pixel point and their directions are calculated. The strain gradient is calculated. For each pixel point, the difference between its principal strain and secondary strain and the principal strain and secondary strain of its surrounding pixel points is calculated and divided by the distance between them. The strain gradient includes two components: the principal strain gradient and the secondary strain gradient. Each component has a magnitude and a direction. Integrate the strain gradient information of all pixel points to generate the strain gradient field distribution map. This map is represented in the form of a vector map, with an arrow at each pixel point. The length of the arrow represents the magnitude of the strain gradient, and the direction of the arrow represents the direction of the strain gradient.

[0114] Based on the strain gradient field distribution map and the interlayer contact area distribution map, calculate the stress at each point within the contact area. According to the interlayer contact area distribution map, determine the grid cells within the contact area. For each grid cell, extract the strain gradient information (from the strain gradient field distribution map) at its four vertices. According to the constitutive relationship (stress-strain relationship) of the material, calculate the stress at each vertex. Assume that the upper material is a linear elastic material, and its constitutive relationship can be expressed by Hooke's law: σ = Eε, where σ represents stress, E represents Young's modulus, and ε represents strain. Young's modulus is determined according to the actual measured value of the upper material. For different materials (such as leather, mesh, rubber), different Young's modulus values are used. The stress tensor includes normal stress (tensile stress or compressive stress) and shear stress (shearing stress). Calculate the principal stress (maximum stress) and secondary stress (minimum stress) at each vertex and their directions. Average the stresses at the four vertices to obtain the average stress of the grid cell. Integrate the average stress information of all grid cells to generate a contact point stress distribution table. This table records the average stress (principal stress and secondary stress) of each grid cell and its direction.

[0115] Based on the contact point stress distribution table, calculate the stress direction vector field within the contact area. For each grid cell in the contact point stress distribution table, extract its average principal stress and secondary stress and their directions. The principal stress direction is the stress direction of the grid cell. Represent the principal stress direction as a unit vector, whose direction is the same as the principal stress direction and the length is 1. Perform the above operations on all grid cells to obtain a stress direction vector field. This vector field describes the stress direction at each point within the contact area.

[0116] According to the stress direction vector field, extract the stress transfer paths within the contact area. Start from the boundary of the contact area and trace the stress transfer paths along the stress direction vector field. At each current point, search for the next point along its stress direction (determined by the stress direction vector field). Select the point closest to the stress direction of the current point as the next point. Add the next point to the stress transfer path. Repeat the above process until reaching the other end boundary of the contact area or no next point can be found. If a stress direction bifurcation is encountered (i.e., there are multiple points with similar directions), select the direction with the maximum stress intensity (principal stress value) to continue tracing. The stress intensity is obtained from the contact point stress distribution table. Record all the traced stress transfer paths. Each path consists of a series of point coordinates. Integrate these paths to generate a stress transfer path set.

[0117] Based on the stress transfer path set and the stress direction vector field, identify the stress characteristic points within the contact area. The stress characteristic points include: stress concentration points, stress dispersion points, and stress turning points. A stress concentration point refers to a point where the stress value (principal stress value) is significantly higher than the surrounding area. It is identified by calculating the stress value difference between each point and its surrounding points. If the difference exceeds a threshold (e.g., twice the standard deviation of the surrounding average stress), then this point is marked as a stress concentration point. A stress dispersion point refers to a point where the stress value is significantly lower than the surrounding area. The identification method is similar to that of stress concentration points, except that the threshold is set to a negative value. A stress turning point refers to a point where the stress direction changes significantly. It is identified by calculating the stress direction difference between each point and its surrounding points. If the difference exceeds a threshold (e.g., 45 degrees), then this point is marked as a stress turning point. Mark all the identified stress characteristic points on the stress direction vector field to generate a stress characteristic point distribution map.

[0118] Based on the stress characteristic point distribution map, measure the delay time of stress transfer under different deformation states. For each stress characteristic point, record its stress value (principal stress value) and stress direction under different deformation states (uniquely determined by the bending angle and pressure value). Analyze the changes in stress value and stress direction with the bending angle and pressure value. Determine the critical deformation states (bending angle and pressure value) where the stress value or stress direction changes significantly (e.g., more than 10% change). Calculate the time difference (difference in bending angle or pressure value) to reach the critical deformation state between different stress characteristic points. This time difference is the delay time of stress transfer. Integrate the delay time information of all stress characteristic points to generate a stress delay coefficient table. This table records the stress transfer delay time between different stress characteristic points.

[0119] Based on the stress delay coefficient table, the stress characteristic point distribution map, and the stress transfer path set, construct a comprehensive stress transfer path map. With the contact area as the background, this map superimposes and displays the following information: the stress transfer path set (using lines of different colors to represent different paths); the stress characteristic point distribution map (using symbols of different shapes to represent different types of characteristic points, such as circles for stress concentration points, triangles for stress dispersion points, and squares for stress turning points); the stress delay coefficient table (annotating numbers on the connections between stress characteristic points to represent the stress transfer delay time). The depth of color of the stress transfer path represents the magnitude of the stress intensity, and the darker the color, the greater the stress intensity. The stress transfer path map clearly shows information such as the path, direction, intensity, and delay time of stress transfer between different layers of the shoe upper.

[0120] Preferably, the calculation of the interlayer coupling coefficient in step S4 includes:

[0121] Conduct stress path hierarchical division according to the stress transfer path map and the three-dimensional hierarchical structure map to obtain a hierarchical stress path set;

[0122] Pair the deformation data for the hierarchical stress path set and the key point deformation characteristic table to obtain an interlayer deformation data pair set;

[0123] Calculate the deformation correlation coefficients for the interlayer deformation data pair set to obtain an interlayer deformation correlation coefficient set;

[0124] Calculate the deformation phase difference data for the interlayer deformation data pair set;

[0125] Calculate the deformation transfer efficiency data for the interlayer deformation data pair set;

[0126] Construct an interlayer coupling coefficient matrix based on the deformation transfer efficiency data, the deformation phase difference data, and the interlayer deformation correlation coefficient set.

[0127] In the embodiment of the present invention, based on the stress transfer path diagram and the three-dimensional hierarchical structure diagram, the stress transfer paths are hierarchically divided. According to the hierarchical relationships of each part (tongue, upper, sole) recorded in the three-dimensional hierarchical structure diagram, determine the layers passed by each stress transfer path. For example, a stress transfer path may start from the sole, pass through the upper, and finally reach the tongue. Divide each stress transfer path into segments according to the layers it passes through. For example, the above path can be divided into a sole segment, an upper segment, and a tongue segment. For each segment, mark the layer (tongue, upper, or sole) to which it belongs. Group all the stress transfer paths by layer to form a hierarchical stress path set. This set contains three subsets: the tongue stress path set, the upper stress path set, and the sole stress path set. Each subset contains all the stress transfer path segments starting from that layer.

[0128] Based on the hierarchical stress path set and the key point deformation characteristic table, pair the deformation data. For each stress transfer path segment in the hierarchical stress path set, determine its starting point and ending point. According to the positions of the starting point and the ending point, find the key points closest to them in the key point deformation characteristic table. Use these two key points as the representative key points of this stress transfer path segment. Extract the displacement data of these two representative key points under different deformation states (uniquely determined by the bending angle and the pressure value). Pair the displacement data of these two key points to form an interlayer deformation data pair. For example, if a stress transfer path segment starts from point A on the sole and ends at point B on the upper, the key point near point A is ID1, and the key point near point B is ID2, then pair the displacement data of key points ID1 and ID2 under all deformation states. Repeat the above operation for all stress transfer path segments in the hierarchical stress path set to obtain an interlayer deformation data pair set. This set contains the displacement data of the representative key point pairs on the stress transfer paths between all adjacent layers.

[0129] Based on the set of interlayer deformation data pairs, calculate the deformation correlation coefficient between adjacent layers. For each data pair in the set of interlayer deformation data pairs, calculate the Pearson correlation coefficient between the displacement data of two key points. The displacement data can be a three-dimensional displacement vector (x, y, z components) or the magnitude of the displacement vector. Calculate the correlation coefficients in the x, y, and z directions respectively, as well as the correlation coefficient of the displacement magnitude. The value range of the correlation coefficient is from -1 to 1, and the larger the absolute value, the stronger the correlation. A positive value indicates a positive correlation, and a negative value indicates a negative correlation. Repeat the above operations for all data pairs in the set of interlayer deformation data pairs to obtain the set of interlayer deformation correlation coefficients. This set contains the deformation correlation coefficients (in the x, y, z directions and magnitude) between the representative key point pairs on the stress transfer path between all adjacent layers.

[0130] Based on the set of interlayer deformation data pairs, calculate the phase difference of the deformation between adjacent layers. For each data pair in the set of interlayer deformation data pairs, analyze the curves of the displacement data of two key points changing with the bending angle and pressure value. Compare the positions (bending angle and pressure value) where the peaks and valleys of the two curves appear. Calculate the time difference (difference in bending angle or pressure value) when the two curves reach the peak or valley. This time difference is the deformation phase difference. If the displacement curve of key point A reaches the peak before the displacement curve of key point B, the phase difference is positive; otherwise, the phase difference is negative. Repeat the above operations for all data pairs in the set of interlayer deformation data pairs to obtain the interlayer deformation phase difference data.

[0131] Based on the set of interlayer deformation data pairs, calculate the transfer efficiency of the deformation between adjacent layers. For each data pair in the set of interlayer deformation data pairs, calculate the proportional relationship between the displacement data of two key points. The displacement data can be a three-dimensional displacement vector (x, y, z components) or the magnitude of the displacement vector. Calculate the displacement ratios in the x, y, and z directions respectively, as well as the ratio of the displacement magnitude. The displacement ratio is defined as the ratio of the displacement of the lower-layer key point to the displacement of the upper-layer key point. For example, if the displacement of the upper vamp key point is 2 mm and the displacement of the tongue key point is 1 mm, the displacement transfer efficiency of the tongue is 1 / 2 = 0.5. Repeat the above operations for all data pairs in the set of interlayer deformation data pairs to obtain the interlayer deformation transfer efficiency data.

[0132] Based on the deformation transfer efficiency data, deformation phase difference data, and the set of interlayer deformation correlation coefficients, a comprehensive interlayer coupling coefficient matrix is constructed. This matrix describes the degree of mutual influence between different layers of the shoe upper during the deformation process. The rows and columns of the matrix represent different layers of the shoe upper (tongue, vamp, sole). The element (i, j) of the matrix represents the influence of the i-th layer on the j-th layer. Each element contains three components: correlation coefficient, phase difference, and transfer efficiency. The correlation coefficient represents the degree of correlation between the deformations of the two layers, the phase difference represents the timing relationship between the deformations of the two layers, and the transfer efficiency represents the amplitude ratio of the deformations of the two layers. The corresponding values in the set of interlayer deformation correlation coefficients, interlayer deformation phase difference data, and interlayer deformation transfer efficiency data are filled into the corresponding elements of the matrix. For example, the matrix element (1, 2) (the second column of the first row) represents the influence of the tongue layer on the vamp layer, and its value includes the deformation correlation coefficient, phase difference, and transfer efficiency between the tongue and the vamp. The interlayer coupling coefficient matrix is a 3×3 matrix because there are three layers (tongue, vamp, sole) in the shoe upper in this embodiment. This matrix comprehensively quantifies the coupling relationship between different layers of the shoe upper.

[0133] Preferably, the interlayer deformation response modeling in step S4 includes:

[0134] Establish interlayer constraint equations based on the interlayer coupling coefficient matrix and the interlayer contact area distribution map;

[0135] Conduct deformation influence factor analysis according to the interlayer constraint equations and the set of deformation state profiles to obtain a deformation influence factor table;

[0136] Extract the co-deformation mode according to the deformation influence factor table and the interlayer coupling coefficient matrix;

[0137] Construct an interlayer response function according to the co-deformation mode and the interlayer constraint equations;

[0138] Integrate the interlayer response function, the co-deformation mode, and the deformation influence factor table to obtain an interlayer interactive deformation model.

[0139] In the embodiments of the present invention, based on the interlayer coupling coefficient matrix and the interlayer contact area distribution map, a constraint equation describing the interaction between adjacent layers is established. According to the interlayer contact area distribution map, the contact type between adjacent layers is determined: fixed connection or sliding contact. For fixed connection points (such as suture connections), the displacement continuity condition is set. That is, the displacements of adjacent layers at the fixed connection point must be equal. Let Ui and Uj be the displacement vectors of the ith layer and the jth layer at the fixed connection point respectively, then the constraint equation is: Ui = Uj. For the sliding contact area, a friction model is introduced. It is assumed that the sliding contact area conforms to the Coulomb friction law, that is, the frictional force is proportional to the normal pressure and the direction is opposite to the relative sliding tendency. Let Ni and Nj be the normal pressures of the ith layer and the jth layer in the sliding contact area respectively, μ be the friction coefficient (determined according to experimental measurements or empirical values, such as 0.3 - 0.7), and Ti and Tj be the tangential forces of the ith layer and the jth layer in the sliding contact area respectively, then the constraint equation is: |Ti - Tj| ≤ μ|Ni + Nj|. According to the interlayer coupling coefficient matrix, the deformation compatibility condition is introduced. That is, the deformations between adjacent layers must satisfy certain correlation, phase difference, and transfer efficiency. Let ρij be the deformation correlation coefficient between the ith layer and the jth layer, φij be the phase difference, ηij be the transfer efficiency, and Ui and Uj be the displacement vectors of the ith layer and the jth layer in the contact area respectively, then the constraint equation is: Uj = ηij * Ui + ΔU, where ΔU is a correction term representing the displacement difference caused by the phase difference φij. A geometric compatibility equation is established to ensure that there is no penetration of the deformed surface. That is, physical impossible overlap cannot occur between adjacent two layers. At each contact point, a normal distance constraint is established: di,j >= 0, where di,j is the normal distance between layer i and layer j. The principle of minimum deformation energy is set as a constraint condition. That is, on the premise of satisfying the above constraint conditions, the deformation energy of the entire shoe upper is minimized. The deformation energy includes elastic deformation energy and frictional dissipation energy. Integrate all the above constraint conditions (displacement continuity condition, friction model, deformation compatibility condition, geometric compatibility equation, principle of minimum deformation energy) to form a complete set of interlayer constraint equations. This set of equations describes the interaction and constraint relationship between adjacent layers during the deformation process.

[0140] Based on the interlayer constraint equations and the set of deformation state profiles, analyze the influence of different factors on the deformation of the shoe upper. The control variable method is adopted. The bending angle and the pressure value are changed respectively, while other conditions are kept unchanged, and the changes in the deformation of each layer are observed. The change range of the bending angle is 0° - 45°, with an interval of 5°; the change range of the pressure value is 0 - 200 N, with an interval of 20 N. For each bending angle and pressure value, solve the interlayer constraint equations to obtain the displacements and stresses of each layer in the equilibrium state. Calculate the deformation sensitivity coefficient. For each key point (the key points defined in step S3), calculate the partial derivatives of its displacement with respect to the bending angle and the pressure value. The larger the partial derivative, the more sensitive the deformation of the key point is to the factor. Analyze the response time (deformation hysteresis) of each layer to the external load. It is determined by comparing the times (number of iterations) for different layers to reach the equilibrium state. Calculate the critical deformation point. The critical deformation point refers to the bending angle or pressure value at which the deformation mode undergoes a sudden change. For example, from linear deformation to non-linear deformation, or from elastic deformation to plastic deformation. It is determined by observing the change curves of the deformation parameters (displacement, stress, strain) with respect to the bending angle and the pressure value. Determine the degree of mutual restriction between each layer. It is determined by analyzing the weights of each constraint condition in the interlayer constraint equations. The larger the weight, the stronger the restriction of the constraint condition on the deformation. Construct an influence factor table to record the influence degrees (sensitivity coefficient, response time, critical deformation point, restriction degree) of different factors (bending angle, pressure, interlayer connection type, material properties, etc.) on the deformation of the shoe upper.

[0141] Based on the deformation influence factor table and the interlayer coupling coefficient matrix, extract the collaborative deformation patterns of the shoe upper under different deformation conditions. The collaborative deformation pattern refers to the consistent deformation trend and the mutually coordinated motion pattern among different layers. Analyze the correlation coefficients in the interlayer coupling coefficient matrix. Mark the layer pairs with the absolute value of the correlation coefficient greater than 0.7 as strongly correlated pairs. Identify the dominant deformation pattern and the secondary deformation pattern. The dominant deformation pattern refers to the pattern that contributes the most to the overall deformation, and the secondary deformation pattern refers to the pattern with less contribution. Determine it by calculating the variance contribution rate of the deformation parameters of each layer. The larger the variance contribution rate, the more important the role of this layer in this pattern. Extract the pattern feature vectors. For each dominant deformation pattern, construct a feature vector to describe the relative displacement direction and amplitude of each layer under this pattern. The elements of the feature vector represent the displacement components of each layer in a specific direction. Calculate the pattern stability index. The pattern stability index refers to the degree of consistency of the deformation patterns among different samples. Determine it by calculating the similarity between the feature vectors of different samples. The higher the similarity, the more stable the pattern, and the range is 0 - 1. Analyze the pattern activation conditions. Determine under what external loads (bending angle and pressure value) a specific collaborative deformation pattern will be triggered. Determine it by analyzing the variation curves of the deformation parameters with the bending angle and pressure value, as well as the variation of the interlayer coupling coefficient. Establish a complete collaborative deformation pattern library. This library contains all the identified dominant deformation patterns and secondary deformation patterns, as well as their feature vectors, stability indices, and activation conditions.

[0142] Based on the collaborative deformation patterns and the interlayer constraint equations, construct functions that describe the deformation response relationship between adjacent layers. For each pair of adjacent layers (such as tongue - upper, upper - sole), establish a response function to map the deformation of the upper - layer structure to the deformation of the lower - layer structure. The input of the response function is the deformation parameters (such as displacement, strain) of the upper - layer structure, and the output is the deformation parameters of the lower - layer structure. Use the piece - wise linear interpolation method to construct the response surface. Divide the ranges of the bending angle and pressure value into several intervals. For example, the bending angle is divided into intervals of every 5°, and the pressure value is divided into intervals of every 20N. Assume that the response function is linear within each interval. Calculate the function coefficients. For each interval, solve the coefficients of the linear function within this interval according to the interlayer constraint equations and the collaborative deformation patterns. Use the least - squares method to minimize the error between the function prediction value and the actual observed value (from the deformation state contour set and the key - point deformation characteristic table). Verify the function prediction accuracy. Use the cross - validation method to divide the data set into a training set and a test set. Fit the function coefficients with the training - set data and evaluate the function prediction error with the test - set data. Control the average error to be less than 3%. Optimize the function parameters to reduce the risk of over - fitting. For example, use the regularization method to limit the complexity of the function coefficients. Establish a complete interlayer response function library. This library contains all the response functions between adjacent layers, and each function consists of a set of piece - wise linear functions.

[0143] Integrate the interlayer response function library, co-deformation mode library, and deformation influence factor table to construct a complete interlayer interaction deformation model for the shoe upper. This model can predict the deformation responses of each layer of the shoe upper and the changes in the overall contour based on the input external loads (bending angle and pressure value). Establish a deformation transfer chain. Determine the transfer path of deformation from the initial stressed layer (such as the sole) to the final response layer (such as the tongue) according to the interlayer response function library. Add a time-series response module. Predict the change process of deformation over time based on the response time information in the deformation influence factor table. Develop an interactive prediction interface. This interface allows users to input the bending angle and pressure value and real-time display the deformation results (displacement, stress, strain) of each layer of the shoe upper and the changes in the overall contour. The deformation results are presented in a graphical way. For example, the stress magnitude is represented by color, and the displacement direction is represented by an arrow. Verify the accuracy of the prediction system. Compare the prediction results of the model with the actual measurement results (from the deformation state contour set) and calculate the prediction error. The error metrics include: mean absolute error, root mean square error, and maximum error. Ensure that the prediction error of the model is within an acceptable range (for example, the average error is less than 5%). Integrate the interlayer response function, co-deformation mode, deformation influence factor table, and time-series response module into a unified mathematical model, which is the interlayer interaction deformation model and can realize the prediction of the overall contour deformation of the shoe upper under any bending and extrusion conditions.

[0144] Preferably, the present invention also provides a shoe upper contour detection system based on an image recognition model for performing the shoe upper contour detection method based on an image recognition model as described above. The shoe upper contour detection system based on an image recognition model includes:

[0145] A multi-state contour acquisition module for acquiring the original multi-state images of the shoe upper under different bending and extrusion degrees and performing contour enhancement processing to obtain a deformation state contour set;

[0146] A structural layer separation module for identifying and marking the key parts of the deformation state contour set to obtain a part marking map; accurately positioning the part interface according to the part marking map to obtain an accurate interface map; constructing a three-dimensional hierarchical structure map according to the accurate interface map;

[0147] A key point deformation tracking module for performing local key point deformation tracking according to the three-dimensional hierarchical structure map to obtain multi-dimensional deformation parameters; performing key part deformation analysis according to the multi-dimensional deformation parameter set to obtain a key point deformation characteristic table;

[0148] The interlayer action response modeling module is used to identify the distribution map of interlayer contact areas according to the key point deformation characteristic table and the three-dimensional hierarchical structure atlas; analyze the interlayer stress transfer path of the distribution map of interlayer contact areas to obtain the stress transfer path map; calculate the interlayer coupling coefficient according to the stress transfer path map and the key point deformation characteristic table to obtain the interlayer coupling coefficient matrix; and perform interlayer deformation response modeling according to the interlayer coupling coefficient matrix and the distribution map of interlayer contact areas to obtain the interlayer interaction deformation model.

[0149] Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be encompassed within the present invention.

[0150] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features invented herein.

Claims

1. A method for detecting the upper contour based on an image recognition model, characterized in that, It includes the following steps: Step S1: Collect the original multi-state images of the shoe upper under different bending and extrusion degrees, and perform contour enhancement processing to obtain a deformation state contour set; Step S2: Identify and mark the key parts of the deformation state contour set to obtain a part marking map; accurately locate the part interface according to the part marking map to obtain an accurate interface map; construct a three-dimensional hierarchical structure map according to the accurate interface map; Step S3: Perform local key point deformation tracking according to the three-dimensional hierarchical structure map to obtain multi-dimensional deformation parameters; Analyze the key part deformation according to the multi-dimensional deformation parameter set to obtain a key point deformation characteristic table; Step S4: Identify the interlayer contact area distribution map according to the key point deformation characteristic table and the three-dimensional hierarchical structure map; Analyze the interlayer stress transfer path of the interlayer contact area distribution map to obtain a stress transfer path map; Calculate the interlayer coupling coefficient according to the stress transfer path map and the key point deformation characteristic table to obtain an interlayer coupling coefficient matrix; Build an interlayer deformation response model according to the interlayer coupling coefficient matrix and the interlayer contact area distribution map to obtain an interlayer interaction deformation model.

2. The method for detecting the upper contour based on the image recognition model according to claim 1, wherein Step S1 includes the following steps: Step S11: Design the bending angle and pressure combination parameters for the shoe upper to obtain a deformation state parameter matrix; Step S12: Collect the original multi-state images of the shoe upper under different bending and extrusion degrees according to the deformation state parameter matrix; Step S13: Perform image preprocessing on the original multi-state images to obtain preprocessed images; Step S14: Segment the shoe upper area of the preprocessed images to obtain a shoe upper mask map; Step S15: Extract and refine the contours according to the shoe upper mask map and the preprocessed images to obtain fine contour line data; Step S16: Perform deformation state correlation integration on the fine contour line data and the deformation state parameter matrix to obtain a deformation state contour set.

3. The method for detecting the upper contour based on the image recognition model according to claim 1, wherein Step S2 includes the following steps: Step S21: Extract the material texture features of the deformation state contour set to obtain a shoe upper texture feature map; Step S22: Perform preliminary clustering of the material areas according to the shoe upper texture feature map to obtain an initial material partition map; Step S23: Screen the candidate structure boundary points according to the initial material partition map and the deformation state contour set to obtain a candidate structure boundary set; Step S24: Identify and mark the parts according to the candidate structure boundary set to obtain a part marking map; Step S25: Accurately locate the interface according to the candidate structure boundary set and the part marking map to obtain an accurate interface map; Step S26: Analyze the hierarchical superposition relationship according to the accurate interface map and the part marking map to obtain a hierarchical relationship table; Step S27: Construct a three-dimensional hierarchical structure map according to the hierarchical relationship table, the accurate interface map and the part marking map.

4. The method for detecting the upper contour based on the image recognition model according to claim 3, wherein Step S25 includes the following steps: Step S251: Extract the junction area location map according to the candidate structure boundary set and the part marking map; Step S252: Calculate the orthogonal gradient field of the junction area location map to obtain an enhanced gradient field map; Step S253: Track the gradient peaks of the enhanced gradient field map to obtain a gradient peak chain; Step S254: Sub-pixel edge precise positioning is performed on the enhanced gradient field map according to the gradient peak chain to obtain a sub-pixel precise boundary; Step S255: Breakpoint detection and connection are performed on the sub-pixel precise boundary to obtain a continuous boundary chain; Step S256: Local enhancement of the fuzzy region is performed according to the continuous boundary chain and the part marking map to obtain an accurate interface map.

5. The method for detecting the upper contour based on the image recognition model according to claim 1, wherein, The local key point deformation tracking described in Step S3 includes: Key point recognition and positioning are performed according to the three-dimensional hierarchical structure atlas to obtain a key point coordinate library; Local feature window extraction is performed according to the key point coordinate library and the deformation state contour set to obtain a key point feature description set; Deformation serialization arrangement is performed according to the key point coordinate library and the deformation state contour set to obtain a deformation path sequence; Point-to-point registration tracking is performed according to the deformation path sequence and the key point feature description set to obtain a point tracking mapping table; Non-rigid deformation analysis is performed according to the point tracking mapping table and the deformation path sequence to obtain a non-rigid deformation field description; Deformation parameter quantization calculation is performed according to the non-rigid deformation field description and the key point coordinate library to obtain multi-dimensional deformation parameters.

6. The method for detecting the upper contour based on an image recognition model according to claim 1, wherein The key part deformation analysis described in Step S3 includes: Key part parameters are extracted according to the multi-dimensional deformation parameter set and the non-rigid deformation field description, where the key parts include toe parameters, tongue parameters, and heel parameters; Bending characteristic calculation is performed on the toe parameters to obtain toe bending data; dynamic tracking of the buckling point is performed according to the toe bending data and the toe parameters to obtain a dynamic characteristic table of the buckling point; Tongue folding pattern recognition is performed on the tongue parameters to obtain tongue fold line distribution data; folding depth and angle measurement are performed according to the tongue fold line distribution data to obtain a tongue folding characteristic table; Heel stability evaluation is performed on the heel parameters to obtain a heel stability index table; shape retention rate calculation is performed according to the heel stability index table to obtain shape retention rate distribution data; Inter-part deformation correlation analysis is performed according to the dynamic characteristic table of the buckling point, the tongue folding characteristic table, and the shape retention rate distribution data to obtain deformation transfer chain data; A key point deformation characteristic table is generated according to the deformation transfer chain data and the multi-dimensional deformation parameters.

7. The method for detecting the upper contour based on an image recognition model according to claim 1, characterized in that, The inter-layer stress transfer path analysis described in Step S4 includes: The inter-layer contact area distribution map is recognized according to the key point deformation characteristic table and the three-dimensional hierarchical structure atlas; Strain gradient calculation is performed according to the key point deformation characteristic table to obtain a strain gradient field distribution map; Contact point stress calculation is performed according to the strain gradient field distribution map and the inter-layer contact area distribution map to obtain a contact point stress distribution table; The stress direction vector field is calculated according to the contact point stress distribution table; The stress transfer path set of the contact point stress distribution table is extracted according to the stress direction vector field; Feature point recognition is performed according to the stress transfer path set and the stress direction vector field to obtain a stress feature point distribution map; Stress delay measurement is performed according to the stress feature point distribution map to obtain a stress delay coefficient table; A stress transfer path map is constructed according to the stress delay coefficient table, the stress feature point distribution map, and the stress transfer path set.

8. The method for detecting the upper contour based on an image recognition model according to claim 1, wherein The inter-layer coupling coefficient calculation described in Step S4 includes: Stress path hierarchical division is performed according to the stress transfer path map and the three-dimensional hierarchical structure atlas to obtain a hierarchical stress path set; Pair the deformation data of the layered stress path set and the key point deformation characteristic table to obtain the interlayer deformation data pair set; Calculate the deformation correlation coefficient of the interlayer deformation data pair set to obtain the interlayer deformation correlation coefficient set; Calculate the deformation phase difference data of the interlayer deformation data pair set; Calculate the deformation transfer efficiency data of the interlayer deformation data pair set; Construct an interlayer coupling coefficient matrix based on the deformation transfer efficiency data, deformation phase difference data, and interlayer deformation correlation coefficient set.

9. The method for detecting the upper surface profile based on an image recognition model according to claim 1, wherein, The interlayer deformation response modeling described in step S4 includes: Establish an interlayer constraint equation based on the interlayer coupling coefficient matrix and the interlayer contact area distribution map; Conduct a deformation influence factor analysis based on the interlayer constraint equation and the deformation state contour set to obtain a deformation influence factor table; Extract the cooperative deformation mode based on the deformation influence factor table and the interlayer coupling coefficient matrix; Construct an interlayer response function based on the cooperative deformation mode and the interlayer constraint equation; Integrate the interlayer response function, cooperative deformation mode, and deformation influence factor table to obtain an interlayer interaction deformation model.

10. A shoe upper contour detection system based on an image recognition model, characterized in that, For implementing the upper surface profile detection method based on an image recognition model as described in claim 1, the upper surface profile detection system based on the image recognition model includes: A multi-state contour acquisition module for acquiring the original multi-state images of the upper surface under different bending and extrusion degrees, and performing contour enhancement processing to obtain a deformation state contour set; A structural hierarchy separation module for identifying and marking the key parts of the deformation state contour set to obtain a part marking map; accurately positioning the part interface according to the part marking map to obtain an accurate interface map; constructing a three-dimensional hierarchy structure map according to the accurate interface map; A key point deformation tracking module for performing local key point deformation tracking according to the three-dimensional hierarchy structure map to obtain multi-dimensional deformation parameters; analyzing the key part deformation according to the multi-dimensional deformation parameter set to obtain a key point deformation characteristic table; An interlayer action response modeling module for identifying the interlayer contact area distribution map according to the key point deformation characteristic table and the three-dimensional hierarchy structure map; analyzing the interlayer stress transfer path of the interlayer contact area distribution map to obtain a stress transfer path map; calculating the interlayer coupling coefficient according to the stress transfer path map and the key point deformation characteristic table to obtain an interlayer coupling coefficient matrix; performing interlayer deformation response modeling according to the interlayer coupling coefficient matrix and the interlayer contact area distribution map to obtain an interlayer interaction deformation model.

Images