A method for identifying misalignment of textile patterns

Abstract

The present application provides a method for identifying misalignment of textile patterns, comprising: performing geometric transformation analysis on extracted pattern contour information, calculating the relative positional relationship between various parts of the pattern, establishing a pattern moisture deformation model, and judging whether the pattern is misaligned or deformed by comparing the difference between the current pattern and a reference model; adding a special waterproof mark on the surface of the wet textile based on the corrected pattern to produce a microscopic mark that does not affect the performance of the fabric, and capturing the position of the waterproof mark through an infrared imaging device as an auxiliary reference point for pattern positioning; tracking the motion trajectory of the waterproof fabric based on the position of the waterproof mark on the pattern, calculating the degree of image distortion based on a motion vector field, and performing real-time correction on the image of the waterproof-marked fabric in motion to eliminate the influence of motion blur on waterproof mark recognition.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for identifying misalignment of textile patterns. Background Art

[0002] Quality control of printed and dyed patterns is a critical step in the textile production process. However, when textiles come into contact with water or are exposed to humidity, patterns can easily become misaligned and deformed, posing a challenge to quality inspection. Furthermore, the texture and diverse patterns on textile surfaces complicate identification. Furthermore, the high-speed movement of fabrics on the production line presents obstacles to real-time detection. To address this issue, it is necessary to develop a misalignment recognition technology that can operate stably in humid conditions. This technology should be able to overcome the effects of moisture on pattern morphology and maintain recognition accuracy even when the fabric is in rapid motion. Furthermore, it is crucial to consider how to add reliable waterproof markings without compromising the fabric's original performance and appearance. Another technical challenge is how to achieve rapid segmentation and recognition of large, complex patterns within limited computing resources, while ensuring recognition accuracy and meeting the real-time requirements of the production line. Furthermore, how to effectively integrate the information from the segmented sub-regions to determine the overall pattern misalignment remains an unresolved issue. Summary of the Invention

[0003] The present invention provides a method for identifying misalignment of textile patterns, which mainly includes:

[0004] Preprocessing textile images in a humid environment, extracting pattern features, building a reference model of the pattern's normal form based on a pre-trained dataset, separating the pattern area from the background area in the original image, and obtaining pattern contour information;

[0005] Perform geometric transformation analysis on the extracted pattern contour information, calculate the relative position relationship of each part of the pattern, establish a pattern moisture deformation model, and determine whether the pattern is dislocated or deformed by comparing the current pattern with the reference model;

[0006] If the pattern is found to be misaligned or deformed, a mathematical model of the effect of moisture on the pattern is established based on the textile material characteristics and production process parameters. By simulating the change pattern under different humidity conditions, the detection results are dynamically corrected to eliminate the interference of moisture on recognition accuracy.

[0007] Based on the corrected pattern, special waterproof markers are added to the surface of the wet textile to create microscopic marks that do not affect the fabric's performance. The positions of the waterproof markers are captured using infrared imaging equipment and used as auxiliary reference points for pattern positioning.

[0008] Based on the waterproof mark position of the pattern, the movement trajectory of the waterproof fabric is tracked, the degree of image distortion is calculated according to the motion vector field, and the image of the moving waterproof mark fabric is corrected in real time to eliminate the impact of motion blur on waterproof mark recognition;

[0009] Based on the correction of fast-moving textile images, the textile texture is segmented, and the waterproof mark is used as the seed point to expand to the surrounding area to determine the texture boundary. The large-area pattern is divided into multiple sub-regions. At the same time, feature extraction and dislocation analysis are performed on each sub-region. The analysis results of each sub-region are summarized to evaluate the dislocation recognition status of the overall textile pattern.

[0010] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0011] The present invention discloses a method for identifying pattern misalignment in textiles. The method establishes a reference model by extracting pattern features, separates the pattern from the background, and obtains contour information. A geometric transformation analysis is performed on the contour, a moisture deformation model is established, and pattern misalignment or deformation is determined. Combining material properties and process parameters, a moisture impact model is established for dynamic correction. A waterproof mark is added to the fabric surface, and infrared imaging is used to capture the mark position for auxiliary positioning. The movement trajectory of the waterproof fabric is tracked, and the moving image is corrected in real time. The texture is segmented, and the boundary is determined using the waterproof mark as a seed point, and the sub-region is divided and analyzed. The present invention solves the problem of identifying pattern misalignment in textiles under humid conditions, improves recognition accuracy, and can be applied to fields such as textile quality control. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 The present invention is a flowchart of a method for identifying misalignment of textile patterns.

[0013] Figure 2 Schematic diagram of a textile pattern misalignment recognition method according to the present invention.

[0014] Figure 3 This is another schematic diagram of a textile pattern misalignment recognition method according to the present invention. DETAILED DESCRIPTION

[0015] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] like Figure 1 -3. In this embodiment, a method for identifying misalignment of textile patterns may specifically include:

[0017] Step S101 , pre-processing the textile image in a humid environment, extracting pattern features, establishing a reference model of the normal form of the pattern based on a pre-trained data set, separating the pattern area and the background area from the original image, and obtaining pattern contour information.

[0018] The method comprises obtaining an original image of a textile under a humid environment, applying a dark channel prior dehazing algorithm to the original image to eliminate blurring effects, and obtaining a dehazed image. The dehazed image is subjected to adaptive histogram equalization and guided filtering, and binarized according to the optimal threshold calculated using the OTSU algorithm to obtain a preprocessed binary image. Morphological operations are performed on the preprocessed binary image, including removing noise regions using an opening operation and filling internal holes in the pattern using a closing operation. The pattern contour is extracted using the Canny edge detection algorithm, and the SIFT feature descriptor of the pattern is extracted using the scale-invariant feature transform algorithm. The SIFT features are compared with the model features based on a pre-trained normal morphological reference model, and the cosine similarity between the feature vectors is calculated. If the cosine similarity is higher than a preset threshold, the pattern is determined to be normal, otherwise it is determined to be an abnormal region. The normal pattern region is segmented by region growing, selecting the pixel with the largest pattern edge gradient as the seed point, calculating the Euclidean color distance and gradient direction difference of adjacent pixels, and adding the pixel to the growing region if both are lower than a preset threshold. This process is repeated until no new pixels meet the growing conditions or the preset maximum number of iterations is reached, thereby obtaining the separated pattern region and background region.

[0019] Specifically, a raw image of textiles in a humid environment is acquired. A dark channel prior dehazing algorithm is used to eliminate blurring effects caused by water vapor. Adaptive histogram equalization is used to adjust image contrast. Guided filtering is used to remove image noise while preserving edge information. The OTSU algorithm automatically calculates the optimal threshold for image binarization, resulting in a preprocessed binary image. Morphological operations are then performed on the preprocessed binary image. Opening operations are used to remove small noisy areas, while closing operations are used to fill small holes within the pattern. The Canny edge detection algorithm is used to extract the pattern outline, and the scale-invariant feature transform (SIFT) algorithm is used to extract local feature descriptors of the pattern, resulting in a geometric and texture description of the pattern. The extracted SIFT features are compared with the model features based on a pretrained normal morphological reference model. The cosine similarity between the feature vectors is calculated, and a similarity threshold is set to determine the degree of feature matching. If the match exceeds the threshold, the pattern is considered normal; otherwise, it is marked as an abnormal area. Region growing segmentation is performed on the patterned regions marked as normal. The pixel with the largest pattern edge gradient is selected as the seed point. The Euclidean color distance and gradient direction difference between adjacent pixels are calculated. If both are less than a preset threshold, the pixel is added to the growing region. This continues until no new pixels meet the growing criteria or the maximum number of iterations is reached, resulting in the separated pattern and background regions. After acquiring the original textile image under a humid environment, a dark channel prior dehazing algorithm is first applied to eliminate the effects of water vapor. This algorithm is based on the observation that most local regions have very low intensity in at least one color channel. By estimating atmospheric illumination and transmittance, a clear image can be restored. Next, adaptive histogram equalization is used to adjust image contrast. The image is divided into several 8x8 pixel blocks. Histogram equalization is performed on each block, and the results are merged using bilinear interpolation to avoid blocking artifacts. Denoising is then performed using a guided filter. This filter uses a guidance image to preserve edge information, with a filter radius of 3 and a regularization parameter of 0.1. Finally, the OTSU algorithm automatically calculates the optimal threshold for binarization. This algorithm maximizes the inter-class variance to find the optimal threshold and segment the image into foreground and background. Morphological operations are performed on the resulting binary image, using a 3x3 rectangular structuring element for opening and closing operations to remove small noisy areas and fill small internal holes, respectively. The Canny edge detection algorithm is then applied to extract the pattern contours, with a low threshold of 50 and a high threshold of 150. The SIFT algorithm is then used to extract local feature descriptors. The SIFT algorithm is invariant to scale and rotation. By constructing a scale space, locating key points, assigning orientations, and generating feature descriptors, a 128-dimensional feature vector is generated. The extracted SIFT features are compared with a pre-trained reference model of normal morphology, and the cosine similarity between the feature vectors is calculated. A similarity threshold of 0.8 is set; if the match exceeds this threshold, the pattern is considered normal; otherwise, it is marked as an abnormal area.Finally, region growing segmentation is performed on the normal pattern area. The pixel with the largest pattern edge gradient is selected as the seed point. The Euclidean distance and gradient direction difference of adjacent pixels are calculated. The color distance threshold is set to 10 and the gradient direction difference threshold is set to 30 degrees. If both are less than the preset threshold, the pixel is included in the growth area. The maximum number of iterations is set to 1000, and finally the separated pattern area and background area are obtained.

[0020] In step S102 , geometric transformation analysis is performed on the extracted pattern contour information to calculate the relative position relationship of each part of the pattern, establish a pattern moisture deformation model, and determine whether the pattern is dislocated or deformed by comparing the difference between the current pattern and the reference model.

[0021] Acquire pattern contour information, use Harris corner detection operator to identify feature points in the pattern, and screen out stable feature points according to a preset corner response threshold and non-maximum suppression window size; calculate the Euclidean distance and angle between the stable feature points, construct a geometric structure description of the pattern, and obtain the relative position relationship of each part of the pattern; for the stable feature points, establish an elastic grid deformation model, divide the pattern into grids of preset size, use bicubic interpolation to calculate the displacement of the internal points of the grid, and obtain a nonlinear deformation field; compare the geometric structure description of the current pattern with that of the reference model, use the accelerated robust feature algorithm to extract local feature descriptors, use the nearest neighbor ratio method for feature matching, and if the Euclidean distance difference between the matching point pairs exceeds the preset threshold, it is determined that the area is dislocated; use the thin plate spline interpolation algorithm to correct the dislocated area, select the matching point pairs that have not been deformed as control points, construct a radial basis function, solve the interpolation coefficient, resample the dislocated area, and obtain the corrected pattern contour information.

[0022] Specifically, based on the extracted pattern contour information, the Harris corner detection operator is used to identify feature points within the pattern. The corner response threshold is set to 0.01, and the non-maximum suppression window size is set to 3x3 pixels to screen out stable feature points. The Euclidean distance and angle between feature points are calculated to construct a geometric structural description of the pattern and determine the relative positional relationships between the pattern components. Based on the identified feature points, an elastic mesh deformation model is established, dividing the pattern into a 16x16 grid with each grid node serving as a control point. Bicubic interpolation is used to calculate the displacement of points within the grid to obtain a nonlinear deformation field. By minimizing the objective function of deformation energy and feature point matching error, the optimal deformation parameters are solved, and a pattern deformation model for humid environments is obtained. The geometric structural description of the current pattern is compared with that of the reference model. The Speeded Up Robust Features (SURF) algorithm is used to extract local feature descriptors. Feature matching is performed using the nearest neighbor ratio method with a matching threshold of 0.7. The Euclidean distance difference between matching point pairs is calculated. If the difference exceeds 5 pixels, the area is considered misaligned. Simultaneously, the affine transformation matrix of the matching point pairs is calculated. If the ratio of the singular values ​​of the transformation matrix exceeds 1.5, the pattern is considered deformed. Based on the misalignment and deformation results and combined with the moisture deformation model, the actual degree of deformation of each pattern component is calculated. The deformed regions are corrected using a thin plate spline interpolation algorithm. Undeformed matching point pairs are selected as control points and radial basis functions are constructed. Interpolation coefficients are calculated, and the deformed regions are resampled to obtain the corrected pattern contour information for subsequent pattern recognition and quality assessment. Starting from the pattern contour information, the Harris corner detection operator is used to identify feature points. The corner response threshold is set to 0.01 to ensure sufficient significance of the detected corners. A 3x3 pixel non-maximum suppression window is used to screen for locally optimal corner points, typically yielding 100-200 stable feature points. The Euclidean distances and angles between these feature points are calculated to construct a geometric structure description matrix. The matrix size is n×n, where n is the number of feature points and each element contains the distance and angle between two feature points. Based on these feature points, a 16x16 elastic mesh deformation model with a total of 256 control points was established. Bicubic interpolation was used to calculate the displacement of internal mesh points with an interpolation kernel size of 4x4. By minimizing the objective function of deformation energy and feature point matching error, a gradient descent method was used for iterative optimization with a learning rate of 0.01 and a maximum number of iterations of 1000 to obtain a nonlinear pattern deformation model for humid environments. When comparing the current pattern with the reference model, the SURF algorithm was used to extract local feature descriptors, setting the Hessian matrix threshold to 300 and extracting a 64-dimensional feature vector. Feature matching was performed using the nearest neighbor ratio method with a matching threshold of 0.7, typically yielding 50-100 pairs of matching points. The Euclidean distance difference between pairs of matching points was calculated; if it exceeded 5 pixels, the region was considered misaligned.At the same time, the RANSAC algorithm is used to estimate the affine transformation matrix of the matching point pairs, and the number of iterations is set to 1000, and the inlier threshold is 2 pixels. If the ratio of the singular values ​​of the transformation matrix exceeds 1.5, the pattern is determined to be deformed. According to the dislocation and deformation judgment results, combined with the nonlinear deformation model obtained previously, the actual deformation degree of each part of the pattern is calculated. Finally, the thin plate spline interpolation algorithm is used for correction, and at least 4 matching point pairs that have not been deformed are selected as control points to construct the radial basis function f(r) = r2log(r). Solve the interpolation coefficient, resample the deformed area, and the grid size is twice the original image to obtain the corrected pattern contour information. This process is achieved through iterative optimization, setting the maximum number of iterations to 100 and the convergence threshold to 0.001 pixels.

[0023] In step S103, if the pattern is found to be misaligned or deformed, a mathematical model of the effect of moisture on the pattern is established based on the material properties of the textile and the production process parameters. By simulating the change pattern of the pattern under different humidity conditions, the detection results are dynamically corrected to eliminate the interference of moisture on the recognition accuracy.

[0024] The material's water absorption parameters and ambient humidity values ​​are obtained, and a multi-layer perceptron neural network is used to establish a nonlinear mapping relationship between the material's water absorption, ambient humidity, and the degree of pattern deformation. A mathematical model of the effect of moisture on the pattern is constructed based on the nonlinear mapping relationship. Real-time humidity data is collected using an ambient humidity sensor, and a time series prediction model is constructed in combination with the material's dynamic moisture absorption characteristics. The time series prediction model uses a long-short-term memory network to process the humidity data sequence and predict the change in the material's internal moisture content over time. The predicted moisture content is input into the multi-layer perceptron neural network to calculate the theoretical deformation state of the pattern under the current humidity conditions. The theoretical deformation state is compared with the actual detected pattern misalignment or deformation, and the deviation matrix between the two is calculated. A state-space model is constructed, in which the state variables are the deviations of the pattern's geometric features, and the observed variables are the actual detection results. The Kalman filter algorithm is applied for dynamic estimation and correction, and the process noise covariance matrix and the observation noise covariance matrix are estimated offline using historical data. The pattern geometric features are corrected using the filtered state estimation values; the corrected pattern geometric features are matched with the original reference model, and the affine transformation matrix is ​​calculated using an image registration algorithm based on mutual information; the mutual information objective function is optimized using a gradient descent method, and the transformation parameters are iteratively updated until convergence or the maximum number of iterations is reached.

[0025] Specifically, a mathematical model for the effect of moisture on patterns was constructed based on textile material properties and production process parameters. A multilayer perceptron neural network was used to establish a nonlinear mapping relationship between material water absorption, ambient humidity, and pattern deformation. The network was trained using experimental data. The input layer contained material parameters and humidity values, the hidden layer used a Reluctant Unit (ReLU) activation function, and the output layer predicted the change in pattern geometric features. Real-time humidity data was collected using an ambient humidity sensor, and a time series prediction model was constructed based on the dynamic moisture absorption characteristics of the material. A long short-term memory (LSTM) network was used to process the humidity data series and predict the change in moisture content within the material over time. The predicted moisture content was input into the previously trained multilayer perceptron to calculate the theoretical deformation state of the pattern under the current humidity conditions. The theoretical deformation state was compared with the actual detected pattern misalignment or deformation, and the deviation matrix between the two was calculated. A state-space model was constructed, in which the state variables were the deviations in the pattern geometric features and the observed variables were the actual detection results. A Kalman filter algorithm was applied for dynamic estimation and correction. The process noise covariance matrix and the observation noise covariance matrix were estimated offline using historical data. The filtered state estimates were used to correct the pattern geometric features. The corrected pattern geometric features are matched to the original reference model, and an image registration algorithm based on mutual information is used to calculate the affine transformation matrix. The mutual information objective function is optimized using gradient descent, and the transformation parameters are iteratively updated until convergence or the maximum number of iterations is reached. Finally, the inverse of the obtained affine transformation matrix is ​​applied to the detection results to perform an inverse transformation, eliminating the interference of moisture on pattern recognition accuracy and obtaining a corrected pattern recognition result. To construct a mathematical model for the effect of moisture on patterns, a multilayer perceptron neural network with three hidden layers is used, with each layer containing 64, 32, and 16 neurons, respectively. The input layer receives five material parameters (such as fiber type and fabric density) and a humidity value, and the output layer predicts the deformation and rotation angle of the pattern in the x and y directions. The Adam optimizer is used with a learning rate of 0.001 and trained for 500 epochs on 10,000 samples. The ambient humidity sensor collects data every 5 seconds. An LSTM network with 12 time steps and a hidden layer size of 32 is constructed to predict the change in moisture content within the material over the next hour. The predicted moisture content sequence was input into a multi-layer perceptron to obtain 24 sets of theoretical deformation states. Actual detection used image matching based on SURF features, extracted 200 feature points, and calculated the Euclidean distance deviation from the theoretical state. A 6-dimensional state space model (x, y displacement, rotation angle, and their rate of change) was constructed, and the observed variables were 4-dimensional (x, y displacement, rotation angle, and area change rate). In the Kalman filter, the diagonal elements of the process noise covariance matrix were set to 0.01, and the diagonal elements of the observation noise covariance matrix were set to 0.1. Fifty Kalman filter iterations were performed on each frame to obtain the corrected pattern geometric features.Image registration uses a mutual information-based approach, calculating mutual information using 32×32 pixel image blocks. Affine transformation parameters are optimized using the L-BFGS-B algorithm, with a maximum number of iterations set to 100 and a convergence threshold of 1e-6. Finally, the inverse of the obtained 3×3 affine transformation matrix is ​​applied to the detection results to perform a reverse transformation, yielding the corrected pattern recognition results. This entire process is implemented on a GPU, with a single frame processing time of under 50 milliseconds.

[0026] A mathematical model of the influence of moisture on patterns is constructed based on different fabric material properties and production process parameters. This model is used to quantitatively describe the geometric distortion effect of moisture content changes on fabric surface patterns, so as to predict the distortion degree and distortion direction parameters. While identifying the fabric surface pattern, the recognition result is subjected to real-time geometric correction based on the predicted pattern distortion degree and distortion direction.

[0027] A database of fabric material properties and production process parameters is constructed, and a mapping relationship between moisture content and pattern geometric distortion is established using a fully connected neural network, wherein the fully connected neural network includes an input layer, a hidden layer, and an output layer; a capacitive sensor array is used to collect moisture content data on the fabric surface, and the moisture content data is used as the input of a long-short-term memory network; a moisture content prediction sequence is obtained based on the output of the long-short-term memory network, and the moisture content prediction sequence is input into the fully connected neural network; the output of the fully connected neural network is obtained to obtain a sequence of predicted values ​​of the degree and direction of pattern distortion; a scale-invariant feature conversion algorithm is used to extract actual pattern feature points and descriptors on the fabric surface, and a time-varying thin plate spline transformation matrix is ​​constructed based on the sequence of predicted values ​​of the degree and direction of pattern distortion; if the time-varying thin plate spline transformation matrix is ​​constructed, non-rigid correction is performed on the actual pattern feature points; a non-rigid registration algorithm based on mutual information is used to compare the corrected pattern with a standard pattern; and the transformation parameters are optimized by a gradient descent method to obtain a pattern recognition result.

[0028] Specifically, a database was constructed based on fabric material properties and production process parameters. A five-layer fully connected neural network was used to establish a mapping between moisture content and pattern geometric distortion. The input layer contained 10 material parameters and one moisture content, while the hidden layers consisted of 64, 32, and 16 neurons, respectively, using the Reluctant Unit (ReLU) activation function. The output layer predicted three distortion parameters and two orientation parameters. The network was trained using stochastic gradient descent with a learning rate set to 0. A capacitive sensor array was used to collect real-time moisture content data on the fabric surface, combined with temperature and humidity sensors to obtain environmental parameters. A long short-term memory (LSTM) network with two hidden layers, each with 64 units, was used to predict moisture content trends over the next 30 minutes. The LSTM network consisted of two hidden layers, each with 64 units. The predicted moisture content sequence was input into the trained fully connected neural network to obtain a sequence of predicted values ​​for the pattern distortion degree and orientation. Simultaneously, the scale-invariant feature transform (SIFT) algorithm was used to extract actual pattern feature points and descriptors from the fabric surface. Based on the predicted distortion parameter sequence, a time-varying thin plate spline transformation matrix was constructed, and the extracted SIFT feature points were subjected to real-time non-rigid correction. The correction process employed a fast approximate nearest neighbor algorithm to accelerate feature point matching. The corrected pattern was compared with the standard pattern using a non-rigid registration algorithm based on mutual information. The transformation parameters were optimized using gradient descent with 100 iterations to obtain the final pattern recognition result. During implementation, a database containing 1,000 combinations of fabric materials and process parameters was constructed, with 20,000 samples of distorted data for each combination at 20 different moisture contents. A five-layer fully connected neural network was used, with an input layer of 11 neurons, three hidden layers of 64, 32, and 16 neurons, respectively, and an output layer of 5 neurons. The Adam optimizer was used with a learning rate of 0.001. 500 epochs of training were performed on 16,000 training samples, and a validation set of 4,000 samples was used. An 8x8 array of capacitive sensors, each covering an area of ​​1 cm², was placed on the fabric surface with a sampling frequency of 10 Hz. The ambient temperature and humidity sensor also had a sampling frequency of 1 Hz. An LSTM network received data from the past 10 minutes and predicted the moisture content for the next 30 minutes, updating the prediction every 5 seconds. The SIFT algorithm extracts features on a 1024x1024 pixel image, setting a keypoint response threshold of 0.04 and extracting a maximum of 1000 feature points. A thin plate spline transform uses 20 control points, dynamically updating the transformation matrix based on the predicted distortion parameters. A fast approximate nearest neighbor algorithm uses a kd-tree structure with a maximum search count of 50. A non-rigid registration algorithm calculates mutual information using 32x32 pixel image blocks, using the L-BFGS-B optimizer with a maximum number of 100 iterations and a convergence threshold of 1e-6. The entire processing flow is implemented on a GPU, with an average processing time of 50 ms for a single pattern recognition and correction pass, meeting real-time requirements.The non-rigid registration algorithm can adapt to various fabrics with moisture content ranging from 5% to 40%, with a pattern recognition accuracy of 98% and an average geometric correction error of less than 2 pixels.

[0029] In step S104, based on the corrected pattern, a special waterproof mark is added to the surface of the wet textile to make a microscopic mark that does not affect the performance of the fabric. The position of the waterproof mark is captured by an infrared imaging device as an auxiliary reference point for pattern positioning.

[0030] The geometric information of a microscopic waterproof mark is obtained. This geometric information is generated using AutoCAD software. The size of the microscopic waterproof mark is controlled to be within the range of 50-100 microns and the shape is an irregular hexagon. Based on this geometric information, a nanoscale waterproof material is used to prepare the mark. The waterproof material solution is atomized into droplets with a diameter of 20-30 microns using piezoelectric inkjet technology. The droplets are then sprayed onto the surface of a damp textile using a piezoelectric inkjet printhead with a resolution of 1200 dpi at a spray frequency of 10 kHz to produce the microscopic waterproof mark. The microscopic waterproof mark is imaged using a multispectral imaging system, and its positional information and geometric features are extracted through wavelet transform and morphological operations. A determination is made as to whether this positional information and geometric features match the original pattern. If so, the RANSAC algorithm is used to select matching point pairs. An affine transformation matrix is ​​calculated based on the matching point pairs to establish an enhanced pattern positioning model containing the waterproof mark.

[0031] Specifically, the geometry of the microscopic waterproofing marks was generated using AutoCAD software based on the corrected pattern contour information. The marks were kept within a 50-100 micron size range and designed as irregular hexagons with side lengths in a ratio of 1:1.2:0.8:1:1.1:0.9 to ensure compatibility with the textile fiber structure. The marks were fabricated using nanoscale waterproofing materials. A piezoelectric inkjet technique atomized the waterproofing material solution into droplets 20-30 microns in diameter. These droplets were then sprayed onto the damp textile surface using a piezoelectric inkjet printhead with a resolution of 1200 dpi at a frequency of 10 kHz, creating the microscopic waterproofing marks. The sprayed textile surface was imaged using a multispectral imaging system with a wavelength range of 1-14 microns, encompassing six bands, and a spatial resolution of 10 microns per pixel. Wavelet transforms and morphological operations were used to extract the positional information and geometric features of the waterproofing marks. The extracted waterproof marker position information was aligned with the original pattern. The RANSAC algorithm was used to select matching point pairs and calculate the affine transformation matrix. The number of iterations was set to 1000, and the inlier threshold was set to 2 pixels. An enhanced pattern positioning model containing the waterproof marker was established for subsequent pattern recognition and positioning. In the actual implementation, the corrected pattern contour information was first imported using AutoCAD 2022 software. A parametric design method was then used to generate the geometry of the microscopic waterproof marker. The marker was designed as an irregular hexagon with side lengths set at a ratio of 1:1.2:0.8:1:1.1:0.9, with the longest side being 100 microns and the shortest side being 80 microns. A 10% hydrophobic nanosilica solution in ethanol was used as the waterproofing material. The markers were produced using a Fujifilm Dimatix DMP-2850 piezoelectric inkjet printer. The printhead nozzle diameter was 21.5 microns, the droplet volume was 10 picoliters, and the jetting frequency was set to 10 kHz. The printing resolution was set to 1200 dpi, corresponding to a spacing of approximately 21.2 microns on the textile surface. The markers were imaged using a Headwall Hyperspec V NIR-SWIR multispectral imaging system, covering a spectral range of 1–14 μm and encompassing six bands (1–2.52 μm, 2.5–52 μm, 5–7.52 μm, 7.5–102 μm, 10–122 μm, and 12–142 μm). Spatial resolution was set to 10 μm / pixel, and exposure time was 50 milliseconds. Image processing was performed using MATLAB R2021b software, using the Haar wavelet transform for multiscale analysis to extract edge information with feature scales of 50–100 μm. Morphological operations used a 3x3 pixel structuring element for opening and closing operations to remove noise and connect disconnected edges. Finally, the RANSAC algorithm was used for feature point matching and affine transformation matrix calculation, with 1000 iterations, a 2-pixel inlier threshold, and a minimum consistent set size of 0.1% of the total image pixels.

[0032] When making microscopic marks that do not affect fabric performance, the diffusion and penetration behavior of the marking liquid between fabric fibers are simulated based on the fabric's hygroscopic properties to obtain the optimal marking liquid formula and coating process parameters; based on the influence of the mark size on the fabric's breathability and softness, the Pareto optimal solution of the mark size and fabric performance is solved, and the mark size range that maximizes the fabric's feel and texture while meeting the pattern positioning requirements is obtained.

[0033] The Lucas-Washburn equation was used to simulate the diffusion and penetration behavior of the marking liquid between fabric fibers. The diffusion rate and penetration depth of the liquid between fibers were calculated based on the fabric porosity, fiber diameter, and surface tension. After obtaining the viscosity and surface tension values ​​of the marking liquid, an orthogonal experimental method was used to design a marking liquid formulation experiment, selecting solvent ratio, water-repellent content, and surfactant concentration as influencing factors. The marking morphology was observed by atomic force microscopy and the marking composition was analyzed by Fourier transform infrared spectroscopy to determine the optimal marking liquid formulation. A relationship model between mark size and fabric air permeability and softness was established, and the fabric air permeability and softness indicators were measured using the KES-F system. A multi-objective genetic algorithm was used to solve the Pareto optimal solution set. The population size, number of iterations, crossover probability, and mutation probability were set to obtain the optimal mark size range that met the pattern positioning requirements and fabric performance. Based on the optimal mark size range and formulation, the coating process was simulated using ANSYS Fluent software. The fluid domain was divided using a hexahedral mesh. The number of meshes, coating roller speed, pressure, and temperature were set to optimize the coating process parameters to ensure mark uniformity and adhesion.

[0034] Specifically, the Lucas-Washburn equation was used to simulate the diffusion and penetration behavior of the marking liquid between fabric fibers. Considering the fabric's porous structure, the diffusion velocity and penetration depth of the liquid between fibers were calculated based on the fabric's porosity, fiber diameter, and surface tension, yielding the optimal viscosity and surface tension values ​​for the marking liquid. An orthogonal experimental design was used to design the marking liquid formulation experiment. Solvent ratio, water-repellent content, and surfactant concentration were selected as influencing factors, and a three-factor, three-level L9(3^4) orthogonal table was established. Marking morphology was observed using atomic force microscopy and marking composition was analyzed using Fourier transform infrared spectroscopy to determine the optimal marking liquid formulation. A model was established to determine the relationship between marking size and fabric air permeability and softness. The fabric's air permeability and softness were measured using the KES-F system. A multi-objective genetic algorithm was used to determine the Pareto optimal solution set, with a population size of 200, an iteration count of 1000, a crossover probability of 0.85, and a mutation probability of 0.05. The optimal marking size range that met both pattern positioning requirements and fabric performance was obtained. Based on the optimal marker size range and formulation, the coating process was simulated using ANSYS Fluent software. The fluid domain was divided into hexahedral meshes with a grid size of 1 million. Boundary conditions included coating roller speed, pressure, and temperature. Coating process parameters were optimized to ensure uniformity and strong adhesion, resulting in the final marking fluid formulation and coating process parameters. In practical application, the Lucas-Washburn equation was first used to simulate the diffusion of the marking fluid in the fabric. The fabric porosity was set to 0.7, the fiber diameter was 20 μm, and the surface tension was 30 mN / m. Numerical calculations revealed the optimal viscosity of the marking fluid to be 10 mPa·s and the surface tension to be 25 mN / m. Subsequently, orthogonal experiments were conducted with solvent ratios (water:ethanol) of 7:3, 6:4, and 5:5, water-repellent content of 2%, 3%, and 4%, and surfactant concentrations of 0.1%, 0.2%, and 0.3%. Marking morphology was observed using a Bruker Dimension Icon atomic force microscope with a scanning range of 10 μm × 10 μm and a resolution of 512 × 512 pixels. The marker composition was analyzed using a ThermoScientific Nicoleti S50 Fourier transform infrared spectrometer with a scanning range of 400–4000 cm² and a resolution of 4 cm². Orthogonal experimental analysis determined the optimal formulation to be a solvent ratio of 6:4, a water repellent content of 3%, and a surfactant concentration of 0.2%. Fabric properties were measured using a KES-F system. Air permeability was tested at a differential pressure of 4 mm water column over a 10 cm² test area, while softness was tested within a flexural stiffness range of 0.01–5 gf·cm² / cm². A multi-objective genetic algorithm, with a weight of 0.5 for air permeability and softness, yielded an optimal marker size range of 60–80 microns.Finally, ANSYS Fluent 2021R2 was used to simulate the coating process. The fluid domain size was 10 mm × 5 mm × 2 mm, and the number of grids was 1,236,742. The coating roller speed was set to 0.5 m / s, the pressure was 0.2 MPa, and the temperature was 25°C. Through optimization, the final coating speed was 0.4 m / s, the pressure was 0.18 MPa, and the temperature was 28°C. The coefficient of variation of the marking uniformity was less than 5%, and the adhesion firmness reached level 4 (remained intact after 100 washes).

[0035] Step S105 , based on the waterproof mark position of the pattern, the motion trajectory of the waterproof fabric is tracked, the degree of image distortion is calculated according to the motion vector field, and the image of the waterproof mark fabric in motion is corrected in real time to eliminate the influence of motion blur on waterproof mark recognition.

[0036] A watertight marker image between consecutive frames is acquired. A motion estimation method based on block matching is used to calculate the watertight marker motion vector field between the consecutive frames. The block matching method sets the block size to a preset pixel value, the search range to a preset pixel range, and the matching criterion to normalized cross-correlation. Based on the motion vector field, an image distortion model based on Point Spread Function is constructed. This distortion model estimates the PSF parameters using Zernike moments, and analyzes the image spectral characteristics using Fourier transform to determine the direction and degree of motion blur. The watertight marker image is deblurred using the Lucy-Richardson deconvolution algorithm, with the maximum number of iterations and the convergence threshold set to a preset value. A preliminary corrected image is obtained. The preliminary corrected image is finely registered using the phase correlation method. The phase difference between the reference image and the corrected image is calculated. The displacement field is obtained through inverse Fourier transform, and the image is resampled using bicubic interpolation. The processing flow is implemented on a GPU using the CUDA parallel computing framework, and the single-frame processing time is controlled within the preset time.

[0037] Specifically, a block-matching-based motion estimation method was used to calculate the motion vector field of the waterproof marker between consecutive frames. The block size was set to 16×16 pixels, the search range was 32×32 pixels, and the matching criterion was normalized cross-correlation. The motion trajectory and velocity information of the marker were obtained. Based on the calculated motion vector field, an image distortion model based on the PointSpreadFunction (PSF) was constructed. The PSF parameters were estimated using Zernike moments. The image spectral characteristics were analyzed through Fourier transform to obtain the direction and degree of motion blur and determine the degree of image distortion. Based on the distortion model, the Lucy-Richardson deconvolution algorithm was used to deblur the moving waterproof marker image. The maximum number of iterations was set to 30, and the convergence threshold was set to 0.001. A preliminary corrected image was obtained. The corrected image was finely registered using the phase correlation method. The phase difference between the reference image and the corrected image was calculated. The displacement field was obtained through inverse Fourier transform. Bicubic interpolation was used for image resampling. Combined with least-squares fitting, sub-pixel registration was achieved, with a registration accuracy of 0.1 pixel. The entire processing flow is implemented on a GPU using the CUDA parallel computing framework, with single-frame processing time kept within 10 milliseconds, meeting real-time correction requirements. In practical applications, a block-matching-based motion estimation method is first used to process consecutive frames of 1920×1080 resolution images. The image is divided into 16×16 pixel blocks, and matching is performed within a 32×32 pixel search range. The normalized cross-correlation coefficient is used as the matching criterion, and a threshold of 0.8 is set. Through parallel computing, a processing speed of 100 frames per second is achieved on an NVIDIA RTX3080 GPU. After obtaining the motion vector field, the PSF is estimated using Zernike moments. The coefficients of the first 20 Zernike polynomials are calculated to reconstruct the PSF kernel function. A 2DFFT Fourier transform is used to analyze the spectral energy distribution in the range 0-π / 2 to determine the main motion direction and blur level. The Lucy-Richardson deconvolution algorithm uses an initial estimate of the original image with a Gaussian blur (σ=1.5), a maximum number of iterations of 30, and a convergence threshold of 0.001. In each iteration, CUDA kernels are used to accelerate convolution operations, resulting in an average frame processing time of 5ms. In the phase correlation method, the image is divided into 32×32 blocks, the phase spectrum of each block is calculated, and the displacement field is obtained through inverse Fourier transform. Least squares fitting of quadratic surfaces is used to achieve sub-pixel registration accuracy of 0.1 pixel. Bicubic interpolation uses a 4×4 pixel neighborhood, and the weight coefficients are calculated using a table lookup for accelerated calculation.

[0038] In step S106, based on the correction of the fast-moving textile image, the textile texture is segmented. Using the waterproof marker as a seed point, the texture boundaries are expanded to the surrounding area, and the large pattern is divided into multiple sub-regions. Feature extraction and misalignment analysis are performed on each sub-region. The analysis results are summarized to evaluate the misalignment recognition status of the overall textile pattern.

[0039] A textile image is received, and an adaptive histogram equalization method is used to enhance the image contrast, the method comprising setting a predetermined block size and a contrast limit parameter; based on the enhanced image, motion blur correction is performed using a Richardson-Lucy algorithm, the algorithm being set with a predetermined number of iterations; bilateral filtering is performed on the corrected image, the bilateral filtering comprising setting a predetermined spatial domain standard deviation and a value range standard deviation; an adaptive threshold segmentation method is used to process the filtered image to obtain a binarized image, the method comprising setting a predetermined window size; if a waterproof mark is detected in the binarized image, texture segmentation is performed using a region growing algorithm using the waterproof mark as a seed point, the algorithm comprising: setting a predetermined local entropy threshold as a growth criterion, and determining whether the local entropy exceeds the preset threshold; if the local entropy exceeds the preset threshold, determining the corresponding region as a target texture region.

[0040] Specifically, the rectified textile images were preprocessed using adaptive histogram equalization to enhance image contrast, with a block size of 8×8 pixels and a contrast limit of 0.01. Motion blur correction was performed using the Richardson-Lucy algorithm with 10 iterations. Bilateral filtering was used to remove noise while preserving edge information, with a spatial standard deviation of 3 and a range standard deviation of 0.1. A binary image was obtained using adaptive threshold segmentation with a window size of 51×51 pixels. Texture segmentation was performed using the waterproof markers as seed points using a region growing algorithm. Local entropy was used as the growth criterion, the threshold was set to 1.2 times the local entropy, and the 8-neighborhood connectivity was used to determine the expansion direction. The maximum number of iterations was set to 10% of the total number of image pixels. The segmented texture regions were then subjected to boundary extraction and sub-region division. Texture boundaries were detected using the Canny edge detection algorithm with a low threshold of 50 and a high threshold of 150. Initial segmentation of large-area patterns was performed using the SLIC superpixel algorithm, with the number of superpixels set to 0.1% of the image area and a compactness factor of 10. The final subregion segmentation results were obtained using a region similarity merging method, with a similarity threshold of 0.85. Feature extraction and misalignment analysis were performed on each subregion. Local binary pattern (LBP) was used to extract texture features, and feature similarity between subregions was calculated. The chi-squared distance was used to measure the differences in LBP histograms. A similarity threshold of 0.9 was set to determine whether a subregion was misaligned. A global Fourier descriptor was calculated to assess overall misalignment, with the first 20 low-frequency coefficients used as feature vectors. Finally, the analysis results for each subregion were summarized, and the overall misalignment recognition of the textile pattern was evaluated by combining local and global features. The entire processing flow was implemented on a GPU using CUDA parallel computing, with a single-frame processing time of less than 20 milliseconds. In practical applications, a 2048×1536 resolution textile image was first preprocessed. Adaptive histogram equalization was performed using 8×8 pixel blocks with a contrast limit parameter of 0.01. Parallel processing was achieved using a CUDA kernel function, taking approximately 2 milliseconds. The Richardson-Lucy algorithm performs motion blur correction, assuming a motion blur kernel size of 21×21 pixels. Iterates 10 times, parallelized on a GPU, and takes approximately 5ms. Bilateral filtering uses a 5×5 pixel window with a spatial domain of σ=3 and a range of σ=0.1, accelerated by a lookup table, and takes approximately 1ms. Adaptive threshold segmentation uses a 51×51 pixel sliding window, calculating the local mean and standard deviation, with the threshold set at the mean minus 0.2 times the standard deviation. It takes approximately 2ms. The region growing algorithm uses the center point of the detected waterproof marker as a seed, calculating the local entropy of a 9×9 neighborhood. The growing threshold is set at 1.2 times the local entropy, with a maximum of 200,000 iterations (approximately 10% of the total number of pixels), and takes approximately 3ms.Canny edge detection uses a 3×3 Sobel operator to calculate gradients, with a low threshold of 50 and a high threshold of 150. Edges are detected through non-maximum suppression and double-threshold linking, taking approximately 1ms. The SLIC superpixel algorithm sets the initial number of superpixels to 3,000 (approximately 0.1% of the image area), iterates 10 times, and uses a 5-dimensional feature vector (Lab* color space and xy coordinates), taking approximately 4ms. Region merging is based on color histogram similarity with a threshold of 0.85, using a union-find data structure to accelerate the merging process, taking approximately 1ms. LBP feature extraction uses a 3×3 neighborhood, 8-bit encoding, and calculates a 256-dimensional histogram. The LBP features of each subregion are normalized, taking approximately 0.5ms. Fourier descriptor calculation uses a 2DFFT, selecting 20 low-frequency coefficients as global features, taking approximately 0.5ms. The entire process is implemented on an NVIDIA RTX 3080 GPU, taking approximately 20ms, meeting real-time processing requirements.

[0041] Obviously, those skilled in the art may make various changes and modifications to the embodiments of the present application without departing from the spirit and scope of the embodiments of the present application. Thus, if these modifications and variations of the embodiments of the present application fall within the scope of the claims of the present application and their equivalents, the present application also intends to include such modifications and variations.

Claims

1. A method for identifying misalignment of textile patterns, characterized in that: The method comprises: pre-processing a textile image under a humid environment, extracting pattern features, establishing a reference model of the normal form of the pattern based on a pre-trained data set, separating the pattern area and the background area from the original image, and obtaining pattern contour information; performing geometric transformation analysis on the extracted pattern contour information, calculating the relative position relationship of each part of the pattern, establishing a pattern moisture deformation model, and judging whether the pattern is dislocated or deformed by comparing the difference between the current pattern and the reference model; if the pattern is dislocated or deformed, establishing a mathematical model of the influence of moisture on the pattern based on the material properties of the textile and the production process parameters, and dynamically correcting the detection result by simulating the change law of the pattern under different humidity conditions to eliminate the interference of moisture on the recognition accuracy; based on the correction After the pattern is formed, special waterproof markers are added to the surface of wet textiles to make microscopic marks that do not affect the performance of the fabric. The position of the waterproof marker is captured by infrared imaging equipment and used as an auxiliary reference point for pattern positioning; based on the position of the waterproof marker in the pattern, the movement trajectory of the waterproof fabric is tracked, the degree of image distortion is calculated according to the motion vector field, and the image of the moving waterproof marker fabric is corrected in real time to eliminate the influence of motion blur on the recognition of waterproof markers; based on the correction of the fast-moving textile image, the textile texture is segmented, and the waterproof marker is used as the seed point to expand to the surrounding area to determine the texture boundary, and the large-area pattern is divided into multiple sub-areas. At the same time, feature extraction and dislocation analysis are performed on each sub-area. The analysis results of each sub-area are summarized to evaluate the dislocation recognition status of the overall textile pattern.

2. The method according to claim 1, wherein The method preprocesses the textile image under a humid environment, extracts pattern features, establishes a reference model of the normal form of the pattern based on a pre-trained data set, separates the pattern area and the background area from the original image, and obtains pattern contour information, including: obtaining the original image of the textile under a humid environment, using a dark channel prior defogging algorithm to eliminate the blur effect on the original image to obtain a defogged image; performing adaptive histogram equalization and guided filtering on the defogged image, calculating the optimal threshold according to the OTSU algorithm for binarization to obtain a preprocessed binary image; performing morphological operations on the preprocessed binary image, using an opening operation to remove noise areas, using a closing operation to fill holes in the pattern, and performing a Canny edge detection algorithm on the preprocessed binary image. Extract the pattern outline and use the scale-invariant feature transform algorithm to extract the SIFT feature descriptor of the pattern; based on the pre-trained reference model of the normal form of the pattern, compare the SIFT feature with the reference model feature of the normal form of the pattern, and calculate the cosine similarity between the feature vectors; if the cosine similarity is higher than the preset threshold, it is determined to be a normal pattern, otherwise it is determined to be an abnormal area; perform region growing segmentation on the area of ​​the normal pattern, select the pixel point with the largest pattern edge gradient as the seed point, calculate the Euclidean distance and gradient direction difference of the adjacent pixels, and if both are less than the preset threshold, include the pixel in the growth area until no new pixel meets the growth condition or the preset maximum number of iterations is reached, thereby obtaining the separated pattern area and background area.

3. The method according to claim 1, wherein The method performs geometric transformation analysis on the extracted pattern contour information, calculates the relative position relationship of each part of the pattern, establishes a pattern moisture deformation model, and judges whether the pattern is dislocated or deformed by comparing the difference between the current pattern and the reference model, including: obtaining pattern contour information, using Harris corner detection operator to identify feature points in the pattern, screening out stable feature points according to a preset corner response threshold and non-maximum suppression window size; calculating the Euclidean distance and angle between the stable feature points, constructing a geometric structure description of the pattern, and obtaining the relative position relationship of each part of the pattern; establishing an elastic grid deformation model for the stable feature points. A shape model is used to divide the pattern into grids of a preset size, and the displacement of the internal points of the grid is calculated using bicubic interpolation to obtain a nonlinear deformation field. The geometric structure description of the current pattern is compared with that of the reference model, and the accelerated robust feature algorithm is used to extract local feature descriptors. The nearest neighbor ratio method is used for feature matching. If the Euclidean distance difference between the matching point pairs exceeds a preset threshold, it is determined that the area is dislocated. The thin plate spline interpolation algorithm is used to correct the dislocated area, and the matching point pairs that have not been deformed are selected as control points. A radial basis function is constructed, and the interpolation coefficient is solved. The dislocated area is resampled to obtain the corrected pattern contour information.

4. The method according to claim 1, wherein If the pattern is dislocated or deformed, a mathematical model of the influence of moisture on the pattern is established according to the material properties of the textile and the production process parameters. By simulating the change law of the pattern under different humidity conditions, the detection results are dynamically corrected to eliminate the interference of moisture on the recognition accuracy, including: obtaining the material water absorption rate parameters and the ambient humidity value, and using a multi-layer perceptron neural network to establish a nonlinear mapping relationship between the material water absorption rate and the ambient humidity and the degree of pattern deformation; constructing a mathematical model of the influence of moisture on the pattern according to the nonlinear mapping relationship; using an ambient humidity sensor to collect real-time humidity data, and combining the dynamic moisture absorption characteristics of the material to construct a time series prediction model; the time series prediction model uses a long short-term memory network to process the humidity data sequence, and predict the change of the moisture content inside the material over time; the predicted moisture content is input into the multi-layer perceptron neural network to calculate the theoretical deformation state of the pattern under the current humidity conditions; comparing the theoretical deformation state with the actually detected pattern dislocation or deformation, and calculating the difference between the two. The invention relates to a method for determining the deviation matrix between the two groups of images; constructing a state space model, wherein the state variable is the deviation of the pattern geometric features and the observation variable is the actual detection result; applying the Kalman filter algorithm for dynamic estimation and correction, and obtaining the process noise covariance matrix and the observation noise covariance matrix through offline estimation of historical data; using the filtered state estimation value to correct the pattern geometric features; matching the corrected pattern geometric features with the original reference model, and calculating the affine transformation matrix using the image registration algorithm based on mutual information; optimizing the mutual information objective function using the gradient descent method, and iteratively updating the transformation parameters until convergence or the maximum number of iterations is reached; and also including: constructing a mathematical model of the influence of moisture on the pattern for different fabric material properties and production process parameters, the model is used to quantitatively describe the geometric distortion effect of the fabric surface pattern caused by the change of moisture content, so as to predict the distortion degree and distortion direction parameters, identify the fabric surface pattern, and perform real-time geometric correction on the recognition result according to the predicted pattern distortion degree and distortion direction.

5. The method according to claim 4, wherein The method constructs a mathematical model of the effect of moisture on patterns based on different fabric material properties and production process parameters. The model is used to quantitatively describe the geometric distortion effect of moisture content changes on fabric surface patterns, so as to predict the distortion degree and distortion direction parameters. While identifying the fabric surface pattern, the recognition result is subjected to real-time geometric correction based on the predicted pattern distortion degree and distortion direction. The method includes: constructing a database of fabric material properties and production process parameters, using a fully connected neural network to establish a mapping relationship between moisture content and pattern geometric distortion, the fully connected neural network including an input layer, a hidden layer and an output layer; using a capacitive sensor array to collect fabric surface moisture content data, and the moisture content data is used as a long-term and short-term The invention relates to an input of a long short-term memory network; a moisture content prediction sequence is obtained according to the output of the long short-term memory network, and the moisture content prediction sequence is input into the fully connected neural network; the output of the fully connected neural network is obtained to obtain a prediction value sequence of the pattern distortion degree and direction; a scale-invariant feature conversion algorithm is used to extract the actual pattern feature points and descriptors on the fabric surface, and a time-varying thin plate spline transformation matrix is ​​constructed according to the prediction value sequence of the pattern distortion degree and direction; if the time-varying thin plate spline transformation matrix is ​​constructed, the actual pattern feature points are non-rigidly corrected; a non-rigid registration algorithm based on mutual information is used to compare the corrected pattern with the standard pattern; the transformation parameters are optimized by the gradient descent method to obtain the pattern recognition result.

6. The method according to claim 1, wherein The method adds a special waterproof mark on the surface of the wet textile based on the corrected pattern to produce a microscopic mark that does not affect the performance of the fabric, and captures the position of the waterproof mark by infrared imaging equipment as an auxiliary reference point for pattern positioning, including: obtaining geometric information of the microscopic waterproof mark, the geometric information is generated by AutoCAD software, the size of the microscopic waterproof mark is controlled within the range of 50-100 microns, and the shape is an irregular hexagon; according to the geometric information, a nano-scale waterproof material is used to prepare the mark, and a piezoelectric inkjet technology is used to atomize the waterproof material solution into droplets with a diameter of 20-30 microns; using the droplets, a piezoelectric inkjet print head with a resolution of 1200dpi is used to position and spray on the surface of the wet textile, and the spray frequency is set to 10kHz to obtain the microscopic waterproof mark; Water marking, using a multispectral imaging system for imaging, extracting the position information and geometric features of the waterproof mark through wavelet transform and morphological operations; judging whether the position information and geometric features match the original pattern, and if so, using the RANSAC algorithm to screen matching point pairs; calculating the affine transformation matrix based on the matching point pairs, and establishing an enhanced pattern positioning model including the waterproof mark; also including: when making microscopic marks that do not affect the performance of the fabric, simulating the diffusion and penetration behavior of the marking liquid between the fabric fibers based on the hygroscopic properties of the fabric, and obtaining the optimal marking liquid formula and coating process parameters; according to the influence of the mark size on the air permeability and softness of the fabric, solving the Pareto optimal solution of the mark size and fabric performance, and obtaining the mark size range that meets the pattern positioning requirements while maximally maintaining the feel and texture of the fabric.

7. The method according to claim 6, wherein: When making the microscopic marking that does not affect the fabric properties, the diffusion and penetration behavior of the marking liquid between the fabric fibers are simulated based on the fabric's hygroscopic properties to obtain the optimal marking liquid formula and coating process parameters; According to the influence of the marking size on the air permeability and softness of the fabric, the Pareto optimal solution of the marking size and fabric performance is solved, and the marking size range that maximizes the feel and texture of the fabric while meeting the pattern positioning requirements is obtained. This includes: using the Lucas-Washburn equation to simulate the diffusion and penetration behavior of the marking liquid between fabric fibers, and calculating the diffusion rate and penetration depth of the liquid between fibers based on the porosity, fiber diameter and surface tension of the fabric; after obtaining the viscosity and surface tension values ​​of the marking liquid, using the orthogonal experimental method to design the marking liquid formula experiment, selecting the solvent ratio, water repellent content and surfactant concentration as influencing factors; observing the marking morphology by atomic force microscopy. The marking components were analyzed using Fourier transform infrared spectroscopy to determine the optimal formulation of the marking liquid; a relationship model between marking size and fabric air permeability and softness was established, and the KES-F system was used to measure the fabric air permeability and softness indicators; a Pareto optimal solution set was solved using a multi-objective genetic algorithm, and the population size, number of iterations, crossover probability, and mutation probability were set to obtain the optimal marking size range that met the pattern positioning requirements and fabric performance; based on the optimal marking size range and formulation, the coating process was simulated using ANSYS Fluent software, and the fluid domain was divided using a hexahedral grid. The number of grids, coating roller speed, pressure, and temperature were set to optimize the coating process parameters to ensure marking uniformity and adhesion firmness.

8. The method according to claim 1, wherein The method includes: obtaining waterproof mark images between consecutive frames, calculating the motion vector field of the waterproof mark between the consecutive frames by using a motion estimation method based on block matching, wherein the block size is set to a preset pixel value, the search range is set to a preset pixel range, and the matching criterion adopts normalized cross-correlation; constructing an image distortion model based on PointSpreadFunction according to the motion vector field, and the distortion model is obtained by Zern The PSF parameters are estimated using Ike moments, and the image spectrum characteristics are analyzed using Fourier transform to obtain the direction and degree of motion blur. The waterproof marker image is deblurred using the Lucy-Richardson deconvolution algorithm, and the deconvolution algorithm sets the maximum number of iterations to a preset value and the convergence threshold to a preset threshold to obtain a preliminary corrected image. The preliminary corrected image is finely aligned using the phase correlation method, and the phase difference between the reference image and the corrected image is calculated. The displacement field is obtained through inverse Fourier transform, and the image is resampled using bicubic interpolation. The processing flow is implemented on a GPU using the CUDA parallel computing framework, and the single-frame processing time is controlled within the preset time.

9. The method according to claim 1, wherein: The correction based on fast-moving textile images segments the textile texture, expands it to the surrounding area with a waterproof mark as a seed point, determines the texture boundary, and divides the large-area pattern into multiple sub-areas; at the same time, performs feature extraction and dislocation analysis on each sub-area, summarizes the analysis results of each sub-area, and evaluates the dislocation recognition status of the overall textile pattern, including: receiving a textile image, using an adaptive histogram equalization method to enhance the image contrast, the method including setting a predetermined block size and contrast limit parameters; based on the enhanced image, using the Richardson-Lucy algorithm to perform motion blur correction, the Richards The on-Lucy algorithm sets a predetermined number of iterations; performs bilateral filtering on the corrected image, wherein the bilateral filtering includes setting a predetermined spatial domain standard deviation and a value range standard deviation; uses an adaptive threshold segmentation method to process the filtered image to obtain a binary image, wherein the method includes setting a predetermined window size; if a waterproof mark is detected in the binary image, the waterproof mark is used as a seed point and a region growing algorithm is used to perform texture segmentation, wherein the algorithm includes: setting a predetermined local entropy threshold as a growth criterion, and judging whether the local entropy exceeds the preset threshold; if the local entropy exceeds the preset threshold, determining that the corresponding area is a target texture area.

Images