Method for textile defect detection based on machine vision

Abstract

The present application relates to the technical field of intelligent quality inspection of the textile industry, and provides a silk flaw detection method based on machine vision, which comprises the following steps: preprocessing a gray image of a silk fabric surface and extracting a multi-fractal feature vector for storage in a dynamic cache pool; generating a flaw category cluster set through adaptive clustering analysis, and determining a typical defect sample based on the mean value of the cluster set; processing the sample through Gaussian-Laplacian filtering to generate a flaw morphology label, and creating a defect heat map sequence and a weaving time sequence stamp based on the same; generating a label-enhanced image by combining the heat map with the original warp and weft image of the fabric, and composing a loom operation cycle defect segment in time sequence; and finally sending the defect segment and the defect type code to a cloth inspection machine terminal. The method realizes efficient flaw detection through multi-scale feature analysis and dynamic clustering, and correlates the production time sequence for quality tracking. The present application can improve detection accuracy and equipment optimization efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent quality inspection technology in the textile industry, and more specifically, to a method for detecting defects in textiles based on machine vision. Background Technology

[0002] In textile production, defect detection is a crucial step in ensuring product quality. Traditional textile defect detection mainly relies on manual visual inspection, which is not only inefficient but also susceptible to human factors, leading to insufficient accuracy and stability of the results. With the development of machine vision technology, machine vision-based textile defect detection methods have gradually gained attention. Machine vision technology acquires image data of the fabric surface through image acquisition devices and then uses image processing algorithms to analyze the images to identify and locate defects. However, existing machine vision inspection methods still have certain limitations when dealing with textiles with complex textures, especially fabrics with delicate textures and sheen variations, such as silk. These methods often struggle to accurately extract the texture features of the fabric, resulting in low defect detection accuracy and insufficient precision in classifying different types of defects.

[0003] In implementing the embodiments of the present invention, the inventors discovered that the prior art has at least the following problems or defects: When processing fabrics with complex textures, the prior art has difficulty in effectively extracting multifractal feature parameters, resulting in insufficient accuracy and reliability of defect detection; when classifying defect categories, the existing clustering analysis methods cannot adaptively adjust according to the texture features of the fabric, thus affecting the accuracy of defect classification; in addition, when generating defect morphology annotations, the prior art lacks an effective multi-scale annotation information generation mechanism, which cannot meet the annotation accuracy requirements under different detection needs. Summary of the Invention

[0004] This invention provides a method, apparatus, and equipment for detecting defects in textiles based on machine vision.

[0005] In a first aspect of the present invention, a machine vision-based method for detecting defects in textiles is provided, comprising:

[0006] The grayscale image data of the fabric surface collected during the production process of silk textiles is preprocessed, multifractal feature parameters are extracted, and the fractal feature vectors are stored in a dynamic cache pool.

[0007] Adaptive clustering analysis is performed on multiple fractal feature vectors in the cache pool to obtain multiple defect category clusters;

[0008] Based on the mean fractal features of each cluster, typical defect samples corresponding to each cluster are determined in the collected fabric texture maps.

[0009] The typical defect samples are processed by Gaussian-Laplace filtering to obtain corresponding defect morphology annotations, and a heat map sequence of defect locations and corresponding weaving time stamps are generated based on the annotations.

[0010] Based on the heat map sequence and the original image of the fabric warp and weft structure, an enhanced image of defect annotation is generated, and the annotated image is composed into a defect segment of the loom operation cycle according to the weaving time stamp.

[0011] The defect fragment and its corresponding defect type code are sent to the fabric inspection machine's quality inspection terminal.

[0012] Furthermore, the determination of typical defect samples based on the mean of fractal features of each cluster includes: selecting the feature with the largest box dimension difference in each cluster.

[0013] Identify the fabric texture primitive features closest to the target feature, and use the matching primitive feature samples as typical defect samples;

[0014] The process of processing typical defect samples to obtain defect morphology annotations includes: using an improved EfficientNet module to process each cluster of samples and generating defect classification confidence scores;

[0015] Multi-scale annotation information is generated based on confidence level and fabric defect feature database.

[0016] Furthermore, the processing of typical defect samples includes: performing defect feature matching retrieval based on fractal feature combination, and determining whether it matches a known defect pattern;

[0017] During matching, a lightweight Ghost convolution module is used to process samples and generate a probability distribution of defect types.

[0018] When no match is found, the samples are processed by a deformable convolutional v2 network to output abnormal defect classification results.

[0019] Furthermore, the preprocessing includes: acquiring industrial camera acquisition parameters and fabric structure specification information;

[0020] Histogram equalization and LOG filtering enhancement are performed on image data to generate multi-scale texture feature maps;

[0021] The process of identifying typical defect samples includes locating yarn breakage areas and abnormal weaving density areas in a multi-scale feature map.

[0022] Furthermore, before sending, the process also includes: generating a loom status diagnostic report based on real-time detection data from the fabric inspection machine;

[0023] The diagnostic report is spatiotemporally aligned and fused with the defect heatmap to generate a production quality correlation map.

[0024] A matrix of equipment tuning suggestions is generated based on a textile process parameter library;

[0025] Send a full-process testing report containing quality profiles and tuning matrices to the terminal.

[0026] Furthermore, the labeled image includes: extracting the warp and weft yarn distribution matrix of the fabric structure repeat unit;

[0027] A weaving defect compensation matrix is ​​generated by combining the defect location matrix and yarn density characteristics.

[0028] The damaged yarn pixels are virtually repaired and rendered using a texture reconstruction algorithm.

[0029] Furthermore, it also includes: establishing a defect level assessment model based on fabric specification standards;

[0030] The defect severity level weight factor is integrated when generating the compensation matrix to achieve visual labeling of defect severity.

[0031] Furthermore, it also includes:

[0032] Based on the defect type encoding, a loom parameter correction scheme is generated by calling the textile process knowledge base;

[0033] Generate warping machine adjustment command sequence based on defect distribution patterns;

[0034] A virtual fabric inspection simulation system was used to simulate the preview image of the repair effect.

[0035] By dynamically coupling and analyzing adjustment commands with equipment operating status, preventative maintenance strategies can be generated.

[0036] In a second aspect of the invention, a machine vision-based textile defect detection device is provided, comprising:

[0037] The image processing module is used to preprocess grayscale image data of the fabric surface collected during the production process of silk textiles, extract multifractal feature parameters, and store the fractal feature vectors in a dynamic cache pool.

[0038] The clustering analysis module is used to perform adaptive clustering analysis on multiple fractal feature vectors in the cache pool to obtain multiple defect category clusters.

[0039] The typical sample determination module is used to determine the typical defect samples corresponding to each cluster in the collected fabric texture map based on the mean of the fractal features of each cluster.

[0040] The annotation generation module is used to process typical defect samples through Gaussian-Laplace filtering to obtain defect morphology annotations, and generate a heat map sequence of defect locations and weaving time stamps based on the annotations;

[0041] The image synthesis module is used to generate enhanced images with defect annotations based on the heat map sequence and the original image of the fabric warp and weft structure, and to assemble the annotated images into defect segments of the loom operation cycle according to the weaving time stamp;

[0042] The data sending module is used to send defect fragments and corresponding defect type codes to the fabric inspection machine's quality inspection terminal.

[0043] In a third aspect of the invention, a computing device is provided, the computing device comprising: at least one processor, a memory, and an input / output unit; wherein the memory is used to store a computer program, and the processor is used to invoke the computer program stored in the memory to perform the method described in any one of the first aspects.

[0044] The embodiments of the present invention have at least the following beneficial effects: Firstly, by extracting multifractal feature parameters and storing them in a dynamic buffer pool, the present invention can effectively address the detection needs of complex textured fabrics such as silk, improving the accuracy and reliability of defect detection. Simultaneously, adaptive clustering analysis can dynamically adjust defect category clusters based on fabric texture features, thereby achieving fine classification of different defect types and improving detection accuracy and adaptability. Furthermore, using Gaussian-Laplace filtering to process typical defect samples and generate multi-scale annotation information can meet the annotation accuracy requirements under different detection scenarios, further enhancing the effect of defect recognition. In the preprocessing stage, histogram equalization and LOG filtering enhancement can generate multi-scale texture feature maps, which helps to more accurately locate yarn breakage areas and weaving density anomaly areas, providing a more reliable image basis for subsequent defect detection.

[0045] Secondly, during the annotation image generation process, this invention extracts the warp and weft yarn distribution matrix of the fabric structure cycle unit and combines it with the defect location matrix and yarn density features to generate a weaving defect compensation matrix. This enables virtual repair rendering of damaged yarn pixels, improving the quality and readability of the annotation image. Simultaneously, based on defect type encoding, it calls the textile process knowledge base to generate a loom parameter correction scheme and combines the defect distribution pattern to generate a warping machine adjustment instruction sequence. This allows for dynamic optimization and preventative maintenance of the textile production process, improving production efficiency and product quality. Furthermore, by simulating a repair effect preview image through a virtual fabric inspection simulation system, the repair effect can be evaluated before actual repair, further optimizing the repair scheme and reducing production costs. Attached Figure Description

[0046] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein:

[0047] Figure 1This is a schematic flowchart of a machine vision-based textile defect detection method provided in an embodiment of the present invention.

[0048] Figure 2 This is a schematic diagram of the structure of a textile defect detection device based on machine vision according to an embodiment of the present invention.

[0049] Figure 3 A schematic diagram of the structure of a computing device according to an embodiment of the present invention is shown. Detailed Implementation

[0050] The principles and spirit of the invention will now be described with reference to several exemplary embodiments. It should be understood that these embodiments are provided merely to enable those skilled in the art to better understand and implement the invention, and are not intended to limit the scope of the invention in any way. Rather, these embodiments are provided to make the invention more thorough and complete, and to fully convey the scope of the invention to those skilled in the art.

[0051] Those skilled in the art will recognize that embodiments of the present invention can be implemented as an apparatus, device, device, method, or computer program product. Therefore, the present invention can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0052] It should be noted that the number of any elements in the accompanying drawings is for illustrative purposes only and not as a limitation, and any naming is for distinction only and has no limiting meaning.

[0053] The following is for reference. Figure 1 , Figure 1 This is a schematic flowchart illustrating a machine vision-based textile defect detection method according to an embodiment of the present invention. Figure 1 As shown, a machine vision-based method for detecting defects in textiles includes:

[0054] S1. Preprocess the grayscale image data of the fabric surface collected during the production process of silk textiles, extract multifractal feature parameters, and store the fractal feature vectors in a dynamic cache pool.

[0055] S2. Perform adaptive clustering analysis on multiple fractal feature vectors in the cache pool to obtain multiple defect category clusters;

[0056] S3. Based on the mean fractal features of each cluster, determine the typical defect samples corresponding to each cluster in the collected fabric texture map.

[0057] S4. The typical defect samples are processed by Gaussian-Laplace filtering to obtain the corresponding defect morphology annotations, and a heat map sequence of defect locations and the corresponding weaving time stamps are generated based on the annotations.

[0058] S5. Generate defect annotation enhanced images based on the heat map sequence and the original image of the fabric warp and weft structure, and assemble the annotated images into defect segments of the loom operation cycle according to the weaving time stamp.

[0059] S6. Send the defect fragment and the corresponding defect type code to the fabric inspection machine quality inspection terminal.

[0060] It should be noted that the preprocessing of grayscale image data of the fabric surface collected during the silk textile production process is mainly to remove noise and enhance the texture features of the image, so as to more accurately extract multifractal feature parameters in subsequent processing. Multifractal feature parameters refer to a set of parameters that reflect the complexity and non-uniformity of the fabric surface texture; these parameters can be obtained through fractal analysis of the grayscale image. Dynamic buffer pool storage refers to storing the extracted fractal feature vectors in a dynamically resizable buffer pool for efficient management and processing of large amounts of data. The buffer pool can dynamically allocate storage space according to actual needs, ensuring fast data retrieval and updates. Adaptive clustering analysis of multiple fractal feature vectors in the buffer pool is to group defects with similar characteristics into one category, forming multiple defect category clusters. Adaptive clustering analysis is an algorithm that can automatically adjust clustering parameters based on data characteristics, making the clustering results more consistent with the actual defect distribution. Based on the mean fractal features of each cluster, typical defect samples corresponding to each cluster are identified in the collected fabric texture maps. This is to screen out the most representative defect samples from a large number of samples, so as to enable more accurate defect morphology labeling in subsequent steps. Typical defect samples refer to samples with typical characteristics of that category within each defect category cluster. These samples can serve as the benchmark for subsequent analysis and processing.

[0061] Specifically, the preprocessing process includes acquiring industrial camera acquisition parameters and fabric structure specifications. Industrial camera acquisition parameters refer to the parameter settings used by the industrial camera when acquiring images, such as exposure time, aperture size, and resolution. These parameters directly affect the quality of the acquired image and the clarity of texture features. Fabric structure specifications refer to the fabric's structure and specifications, such as warp and weft density, yarn thickness, and fabric type. This information is crucial for understanding the fabric's texture features and defect distribution. Histogram equalization and LOG filtering enhancement of the image data aim to enhance image contrast and texture details, making defect features more prominent. Histogram equalization is a commonly used image enhancement method that adjusts the histogram distribution of the image to make the grayscale value distribution more uniform, thereby enhancing image contrast. LOG filtering enhancement is an image enhancement method based on the Laplacian operator, which enhances the image's edges and texture details by applying Gaussian smoothing and Laplacian filtering. Generating multi-scale texture feature maps involves processing the image with different scale parameters to obtain texture feature maps at different scales, in order to better capture the multi-scale features of defects. Multi-scale texture feature maps refer to texture feature images extracted at different scales. These images can contain texture information from macro to micro levels, helping to more comprehensively analyze the texture features and defect distribution of fabrics. When identifying typical defect samples, it is necessary to locate yarn breakage areas and weaving density anomaly areas in the multi-scale feature maps. Yarn breakage areas refer to the locations where yarns break in the fabric; these areas usually exhibit obvious texture changes and defect features. Weaving density anomaly areas refer to areas in the fabric where the weaving density does not conform to normal specifications; these areas may also contain defects. By locating these areas, typical defect samples can be screened more accurately, providing a more reliable basis for subsequent defect morphology annotation.

[0062] Preferably, when performing adaptive clustering analysis on multiple fractal feature vectors in the cache pool, the K-means clustering algorithm can be used. K-means clustering is a commonly used clustering algorithm that divides data points into K clusters, ensuring high similarity among data points within each cluster and low similarity between data points in different clusters. In this embodiment, the value of K can be dynamically adjusted based on the distribution of the fractal feature vectors to obtain the best clustering effect. Specifically, a preliminary analysis of the fractal feature vectors can be performed to determine an initial range of K values, and then the K value can be adjusted iteratively until the stability of the clustering results reaches a preset threshold. When determining typical defect samples based on the fractal feature mean of each cluster, the nearest neighbor search algorithm can be used. The nearest neighbor search algorithm is an algorithm that finds the data point in a dataset that is most similar to the target data. In this embodiment, the fractal feature mean of each cluster can be used as the target data, and then the sample most similar to the target data is searched in the fabric texture map and used as the typical defect sample. Specifically, the Euclidean distance or other similarity metrics between each sample and the target data can be calculated, and the sample with the smallest distance or the highest similarity can be selected as the typical defect sample. When processing typical defect samples to obtain defect morphology annotations, an improved EfficientNet module can be used. EfficientNet is an efficient convolutional neural network model that expands in depth, width, and resolution through a composite scaling method, achieving higher accuracy and efficiency. In this embodiment, the EfficientNet module can be improved to better suit the fabric defect detection task. Specifically, the parameters of the convolutional and pooling layers in the network structure can be adjusted to increase the network's depth and width to better extract defect features. Simultaneously, the network can be pre-trained to achieve better initial performance on the fabric defect dataset, and then fine-tuned to further improve the accuracy of defect morphology annotation.

[0063] In some embodiments, determining typical defect samples based on the mean fractal features of each cluster includes: selecting the feature with the largest box dimension difference in each cluster.

[0064] Identify the fabric texture primitive features closest to the target feature, and use the matching primitive feature samples as typical defect samples;

[0065] The process of processing typical defect samples to obtain defect morphology annotations includes: using an improved EfficientNet module to process each cluster of samples and generating defect classification confidence scores;

[0066] Multi-scale annotation information is generated based on confidence level and fabric defect feature database.

[0067] It should be noted that the process of determining typical defect samples based on the mean fractal features of each cluster is mainly to screen out the most representative defect samples from a large number of samples, so as to enable more accurate defect morphology annotation in the subsequent process. The maximum box-dimensionality difference feature refers to selecting the feature with the largest box-dimensionality difference among each cluster. Box-dimensionality is a fractal feature parameter used to describe the complexity and non-uniformity of an image. By selecting the maximum box-dimensionality difference feature, the typical features of defects in the cluster can be better reflected. Fabric texture primitive features refer to the features of the basic building blocks of fabric texture. These features can be used to match and identify defects. By determining the fabric texture primitive features closest to the target feature, the matching primitive feature samples can be used as typical defect samples. When processing typical defect samples to obtain defect morphology annotations, an improved EfficientNet module is used to process the samples of each cluster to generate defect classification confidence scores. EfficientNet is an efficient convolutional neural network model, and its improvement allows it to better adapt to fabric defect detection tasks. Multi-scale annotation information is generated based on the confidence scores and the fabric defect feature database, which can provide a more accurate basis for subsequent defect identification and classification.

[0068] Specifically, the maximum box dimension difference feature refers to selecting the feature vector with the largest box dimension difference among the fractal feature vectors of the cluster. Box dimension is a method for calculating fractal dimension, which measures the complexity of an image by calculating the number of boxes covering the image at different scales. In this embodiment, the box dimension of all samples in each cluster can be calculated, and the feature vector with the largest box dimension difference can be selected as the target feature. Fabric texture primitive features refer to the features of the basic building blocks of fabric texture, which can include texture orientation, frequency, contrast, etc. When determining typical defect samples, the mean of the fractal features of the cluster can be matched with the fabric texture primitive features, and the primitive feature sample closest to the target feature can be selected. The improved EfficientNet module refers to improving the EfficientNet model to better adapt to the fabric defect detection task. Specifically, the parameters of the convolutional and pooling layers in the network structure can be adjusted to increase the depth and width of the network to better extract defect features. Defect classification confidence refers to the confidence level of the model in each defect classification. The improved EfficientNet module can generate the confidence level for each defect category. The fabric defect feature database is a database containing various fabric defect features, used to generate multi-scale annotation information. Multi-scale annotation information refers to the information that annotates defects at different scales, and this information can include the location, size, shape, etc. of the defects.

[0069] Preferably, when determining the fabric texture primitive feature closest to the target feature, a nearest neighbor search algorithm can be used. Specific steps include: calculating the Euclidean distance or other similarity measure between the mean of the fractal features of the cluster and the fabric texture primitive features, and selecting the primitive feature sample with the smallest distance or the highest similarity as a typical defect sample. The improved EfficientNet module can be constructed through the following steps: First, adjust the network's input layer to receive grayscale image data of fabric defect images; second, increase the network's depth and width by expanding it in depth, width, and resolution using a composite scaling method; finally, pre-train the network to achieve better initial performance on the fabric defect dataset, and then fine-tune it to further improve the accuracy of defect morphology annotation. When generating multi-scale annotation information, annotation information at different scales can be generated based on defect classification confidence and information in the fabric defect feature database. Specifically, corresponding bounding boxes and annotation information can be generated at different scales based on the location and size of the defect; this annotation information can be used for subsequent defect recognition and classification.

[0070] In some embodiments, processing typical defect samples includes: performing defect feature matching retrieval based on fractal feature combination, and determining whether it matches a known defect pattern;

[0071] During matching, a lightweight Ghost convolution module is used to process samples and generate a probability distribution of defect types.

[0072] When no match is found, the samples are processed by a deformable convolutional v2 network to output abnormal defect classification results.

[0073] It's important to note that fractal feature combination refers to combining multiple fractal feature parameters to form a comprehensive feature vector, used to more fully describe the texture features of defects. By performing defect feature matching retrieval through fractal feature combination, it's possible to determine whether a sample matches a known defect pattern. The lightweight Ghost convolution module is an efficient convolution operation module that reduces computation while maintaining model performance. The deformable convolution v2 network is an improved convolutional neural network capable of handling geometric deformations in images, suitable for defect detection in complex textures. These techniques enable accurate defect classification and anomaly detection.

[0074] Specifically, fractal feature combination refers to combining multiple fractal feature parameters (such as box dimension, correlation dimension, etc.) into a feature vector. These feature parameters can be obtained by fractal analysis of grayscale images of fabric surfaces, reflecting the complexity and non-uniformity of fabric texture. During defect feature matching retrieval, the extracted fractal feature combination is compared with a known defect pattern feature library to determine if a matching known defect pattern exists. The lightweight Ghost convolution module is a convolution operation based on the Ghost module, which reduces computation by generating redundant feature maps. In this embodiment, parameters of the Ghost module, such as the ratio of redundant feature map generation, can be set to optimize computational efficiency. The deformable convolution v2 network is an improved convolutional neural network that adapts to geometric deformations in images by introducing deformable convolution kernels. When processing samples, the deformable convolution v2 network can dynamically adjust the shape and size of the convolution kernels according to the texture features of the samples, thereby better capturing the features of defects. Abnormal defect classification results refer to the classification results output by the deformable convolutional v2 network when the sample does not match the known defect pattern, and are used to identify new or unknown defect types.

[0075] Preferably, the construction of fractal feature combinations can be achieved through the following steps: First, extract multiple fractal feature parameters, such as box dimension and correlation dimension, from the grayscale image of the fabric surface; then, normalize these parameters to eliminate dimensional differences between different parameters; finally, combine the normalized parameters into a feature vector. When performing defect feature matching retrieval, a matching threshold can be set. When the similarity between the fractal feature combination and the known defect pattern is higher than this threshold, the match is considered successful. The construction of the lightweight Ghost convolutional module can include the following steps: First, define the redundant feature map generation ratio of the Ghost module, for example, set it to 0.5, indicating that the number of generated redundant feature maps is half the number of original feature maps; then, merge the redundant feature maps with the original feature maps to form a complete feature map output. The construction of the deformable convolutional v2 network can include the following steps: First, define the offset parameters of the deformable convolutional kernel, which can be learned and used to adjust the shape and size of the convolutional kernel; then, apply the adjusted convolutional kernel to the input image to extract feature maps; finally, output the classification results of abnormal defects through subsequent layers of the network (such as fully connected layers). When processing samples, fractal features can be combined as input parameters and processed through a lightweight Ghost convolution module and a deformable convolution v2 network to generate defect type probability distributions or abnormal defect classification results.

[0076] In some embodiments, the preprocessing includes: acquiring industrial camera acquisition parameters and fabric structure specification information;

[0077] Histogram equalization and LOG filtering enhancement are performed on image data to generate multi-scale texture feature maps;

[0078] The process of identifying typical defect samples includes locating yarn breakage areas and abnormal weaving density areas in a multi-scale feature map.

[0079] It's important to note that preprocessing the acquired grayscale image data of the fabric surface is crucial for enhancing the image's texture features, facilitating subsequent extraction of multifractal feature parameters. Preprocessing includes acquiring industrial camera parameters and fabric structure specifications, information essential for understanding the image's background and fabric characteristics. Histogram equalization and LOG filtering enhancement are two commonly used image processing techniques to improve image contrast and texture clarity. Histogram equalization adjusts the image's grayscale distribution, making the contrast more uniform, while LOG filtering enhancement, through a combination of Gaussian smoothing and the Laplacian operator, highlights image edges and texture details. Multi-scale texture feature maps refer to texture feature images generated at different scales, used to capture detailed information about the fabric texture. These preprocessing steps allow for more accurate localization of yarn breakage areas and areas of abnormal weave density, providing a more reliable basis for defect detection.

[0080] Specifically, histogram equalization is an image enhancement technique that adjusts the grayscale histogram of an image to make the distribution of grayscale values ​​more uniform. In this embodiment, histogram statistics can be performed on the acquired grayscale image, and then the original grayscale values ​​can be mapped to new grayscale values ​​through a cumulative distribution function, thereby enhancing the image contrast. LOG filtering enhancement is an image processing method that combines Gaussian smoothing and the Laplacian operator, which can effectively enhance the edge features of an image and further improve the clarity of texture. The parameters of Gaussian smoothing include the standard deviation, which is used to control the degree of smoothing; the parameters of the Laplacian operator include the size and shape of the operator, which are used to detect edges in the image. Multi-scale texture feature maps refer to texture feature images generated at different scales, which can contain texture information from macro to micro. In this embodiment, texture feature maps of different scales can be generated by changing the standard deviation of Gaussian smoothing and the size of the Laplacian operator. Yarn breakage areas refer to the locations where yarns break in a fabric; these areas usually exhibit obvious texture changes and defect features. Abnormal weave density areas refer to regions in a fabric where the weave density does not conform to normal specifications; these areas may also contain defects. By locating these areas, typical defect samples can be screened more accurately, providing a more reliable basis for subsequent defect morphology labeling.

[0081] Preferably, the specific steps of histogram equalization processing may include: first, calculating the histogram of the grayscale image and counting the number of pixels for each grayscale value; then, calculating the cumulative distribution function to map the original grayscale values ​​to a new grayscale value range, for example, mapping the grayscale value range from [0,255] to [0,255], making the image contrast more uniform. The specific steps of LOG filtering enhancement may include: first, selecting a suitable standard deviation (e.g., σ = 1.0) for Gaussian smoothing to remove noise from the image; then, applying a 3×3 or 5×5 Laplacian operator for edge detection to enhance the texture features of the image. When generating multi-scale texture feature maps, multiple different Gaussian smoothing standard deviations (e.g., σ = 0.5, 1.0, 2.0) and Laplacian operator sizes (e.g., 3×3, 5×5) can be set to process the image separately, obtaining texture feature maps at different scales. When locating yarn breakage areas and areas of abnormal weaving density, these areas can be identified by analyzing the grayscale changes and texture patterns in the multi-scale texture feature maps and setting thresholds. For example, a grayscale variation threshold can be set. When the grayscale variation of a certain area exceeds this threshold, it is considered that the area may have yarn breakage or abnormal weaving density. Through these refined steps, images can be preprocessed more accurately, providing higher-quality input data for subsequent defect detection.

[0082] In some embodiments, the method further includes generating a loom status diagnostic report based on real-time detection data from the fabric inspection machine before sending;

[0083] The diagnostic report is spatiotemporally aligned and fused with the defect heatmap to generate a production quality correlation map.

[0084] A matrix of equipment tuning suggestions is generated based on a textile process parameter library;

[0085] Send a full-process testing report containing quality profiles and tuning matrices to the terminal.

[0086] It should be noted that the loom status diagnostic report is generated by analyzing real-time detection data from the fabric inspection machine before sending defect fragments and their corresponding defect type codes. This diagnostic report reflects the loom's operating status during production, including the presence of factors affecting fabric quality such as abnormal vibrations and tension changes. Spatiotemporal alignment and fusion refers to matching the temporal information in the diagnostic report with the spatial information in the defect heatmap to generate a map that intuitively reflects the correlation between quality problems and equipment status during production. The production quality correlation map is a visualization tool that combines the distribution of fabric defects with the loom's operating parameters, helping operators quickly locate problems and take corrective action. The equipment optimization suggestion matrix is ​​generated based on a textile process parameter library, providing operators with specific equipment adjustment suggestions for current quality problems to optimize the production process and improve product quality.

[0087] Specifically, real-time detection data from the fabric inspection machine refers to data related to fabric quality and loom operating status collected in real time by the inspection machine through sensors and other equipment during the fabric production process. This data may include the location, density, and type of fabric defects, as well as parameters such as loom speed, tension, and temperature. The loom status diagnostic report is generated by analyzing this real-time data, providing information on the loom's operating status, such as the presence of abnormal vibrations or tension fluctuations. Spatiotemporal alignment and fusion refers to matching the timestamps in the diagnostic report with the location information in the defect heatmap to ensure consistency in time and space. The defect heatmap is a visual representation that uses color intensity to indicate the distribution density and severity of defects. The production quality correlation map is a visualization tool that combines the distribution of fabric defects with the loom's operating parameters, helping operators quickly identify the correlation between quality problems and equipment status. The equipment tuning suggestion matrix is ​​generated based on a textile process parameter library and contains specific equipment adjustment suggestions for different quality problems, such as adjusting loom tension and speed parameters to optimize the production process and improve product quality.

[0088] Preferably, the specific steps for generating a loom status diagnostic report may include: first, extracting parameters related to the loom's operating status, such as rotational speed, tension, and temperature, from real-time monitoring data from the fabric inspection machine; then, analyzing these parameters to determine if there are any abnormal fluctuations or deviations from the normal range; finally, generating a diagnostic report containing these analysis results. The specific process of spatiotemporal alignment and fusion may include: first, obtaining the timestamp and location information from the defect heatmap; then, matching the time information in the diagnostic report with the timestamp of the defect heatmap to ensure temporal consistency; next, spatially associating the equipment status information in the diagnostic report with the location information in the defect heatmap to generate a production quality correlation map. The generation of the equipment optimization suggestion matrix can be based on a textile process parameter library, which contains optimal process parameters corresponding to various fabric types and quality problems. The specific steps may include: first, retrieving relevant optimization suggestions from the process parameter library based on the type and severity of the current quality problem; then, integrating these suggestions into a matrix for quick reference and application by operators. For example, if yarn breakage is detected in the fabric, the tuning suggestion matrix may recommend reducing the loom speed or adjusting the parameters of the tension controller to reduce the risk of yarn breakage.

[0089] In some embodiments, the labeled image includes: extracting the warp and weft yarn distribution matrix of the fabric weave repeat unit;

[0090] A weaving defect compensation matrix is ​​generated by combining the defect location matrix and yarn density characteristics.

[0091] The damaged yarn pixels are virtually repaired and rendered using a texture reconstruction algorithm.

[0092] It should be noted that the process of generating the labeled image involves extracting the warp and weft yarn distribution matrix of the fabric's weave cycle units and combining it with the defect location matrix and yarn density features to generate a weaving defect compensation matrix. This process aims to virtually repair and render damaged yarn pixels using texture reconstruction algorithms, thereby generating a labeled image that more closely resembles the actual state of the fabric. The warp and weft yarn distribution matrix is ​​a matrix describing the distribution of warp and weft yarns in the fabric's weave structure, reflecting the fabric's texture features. The defect location matrix records the specific location information of defects in the fabric, while the yarn density features describe the density of the yarns. The weaving defect compensation matrix generated using this information provides the necessary data support for the texture reconstruction algorithm, enabling the virtual repair and rendering of damaged yarn pixels.

[0093] Specifically, the fabric weave cycle unit refers to the basic unit that repeats in the fabric structure, determining the fabric's texture characteristics. The warp and weft yarn distribution matrix is ​​a two-dimensional matrix where the element values ​​represent the distribution of warp and weft yarns in the fabric; for example, the values ​​in the matrix can represent the interlacing pattern or density of the yarns. The defect location matrix is ​​a matrix recording the locations of defects, with each element value corresponding to the coordinate position of the defect in the fabric. The yarn density feature is a parameter describing the density of yarns; it can be a scalar or vector, representing the quantity or distribution density of yarns per unit area. The weaving defect compensation matrix is ​​generated based on the above information; it contains compensation data for fabric defects, guiding the texture reconstruction algorithm to repair damaged yarn pixels. The texture reconstruction algorithm is an image processing technique that can virtually repair damaged areas based on the fabric's texture characteristics and the compensation matrix, making the repaired image closer to the real fabric texture. This process involves the analysis and simulation of the fabric texture, as well as the estimation and filling of pixel values ​​in the damaged areas.

[0094] Preferably, extracting the warp and weft yarn distribution matrix of the fabric weave cycle unit can be achieved through the following steps: First, preprocess the fabric image to enhance texture features; then, identify the boundaries and internal structure of the fabric weave cycle unit using image analysis techniques; finally, generate a two-dimensional matrix where the element values ​​represent the distribution of warp and weft yarns. For example, edge detection algorithms can be used to identify yarn boundaries, and the yarn distribution density can be calculated using statistical methods. When generating the weaving defect compensation matrix, the compensation value for each defect location can be calculated based on the defect location matrix and yarn density features. For example, if the yarn density at a certain defect location is low, the compensation value at that location can be appropriately increased to achieve a better repair effect. The specific implementation of the texture reconstruction algorithm can include: First, determining the repair strategy for the damaged area based on the fabric texture features and the compensation matrix; then, estimating and filling the pixel values ​​of the damaged area using interpolation or texture synthesis techniques. For example, interpolation methods based on neighborhood information can be used, or a texture synthesis algorithm can be employed to copy similar texture patterns from other areas of the fabric to the damaged area. During this process, parameters of the algorithm, such as interpolation weights or texture block size, can be set to optimize the repair effect.

[0095] In some embodiments, the method further includes: establishing a defect level assessment model based on fabric specification standards;

[0096] The defect severity level weight factor is integrated when generating the compensation matrix to achieve visual labeling of defect severity.

[0097] It should be noted that a defect grading assessment model was introduced during the image annotation process. This model is used to quantitatively assess the severity of fabric defects. By integrating defect grading weighting factors into the weaving defect compensation matrix, the severity of defects can be visualized, thus more intuitively reflecting the impact of defects on product quality. The defect grading assessment model is a quantitative tool based on fabric specification standards. It can classify defects according to factors such as type, size, and location. When generating the weaving defect compensation matrix, the defect location matrix and yarn density characteristics are combined, while also considering defect grading weighting factors, to ensure that the compensation matrix can more accurately reflect the severity of defects, providing more reasonable data support for subsequent texture reconstruction and repair.

[0098] Specifically, the defect level assessment model is established based on the fabric's specifications and quality requirements, defining the level standards corresponding to different types of defects. For example, defects can be classified into several levels, such as minor, moderate, and severe, based on factors such as size, shape, and color differences. Each level has a corresponding weighting factor used to quantify the severity of the defect. When generating the weaving defect compensation matrix, it is first necessary to determine the defect location matrix, which records the specific location information of defects in the fabric; then, combined with yarn density characteristics, the yarn density at each defect location is evaluated; finally, the defect level weighting factor is incorporated into the calculation of the compensation matrix. For example, for severe defects, a higher weighting factor can be assigned, thus reflecting their greater impact in the compensation matrix. The generation process of the weaving defect compensation matrix involves a comprehensive consideration of defect location, yarn density, and defect level. By calculating the compensation value for each defect location, it provides a basis for the texture reconstruction algorithm to repair the defect.

[0099] Preferably, the construction of the defect level assessment model may include the following steps: First, collect different types of fabric defect samples and classify and grade these samples according to fabric specification standards; then, define the quantification standard for each level, for example, assign weight factors to each level based on parameters such as the proportion of defect area to fabric area and the difference between defect color and surrounding fabric color. When generating the weaving defect compensation matrix, the calculation method for the compensation value can be specifically set. For example, for each defect location, the compensation value is calculated based on its yarn density and defect level weight factor. If the yarn density at a defect location is low and the defect level is high, the compensation value at that location can be appropriately increased to ensure that the repaired fabric texture is closer to the real state. When processing damaged yarn pixels, the texture reconstruction algorithm can adjust the repair strategy based on the compensation value in the compensation matrix. For example, for areas with high compensation values, a more refined repair algorithm can be used to improve the repair effect. Through these refined steps, the severity of defects can be assessed more accurately, and higher-quality annotated images can be generated.

[0100] In some embodiments, it also includes:

[0101] Based on the defect type encoding, a loom parameter correction scheme is generated by calling the textile process knowledge base;

[0102] Generate warping machine adjustment command sequence based on defect distribution patterns;

[0103] A virtual fabric inspection simulation system was used to simulate the preview image of the repair effect.

[0104] By dynamically coupling and analyzing adjustment commands with equipment operating status, preventative maintenance strategies can be generated.

[0105] Based on defect type coding, a textile process knowledge base is invoked to generate loom parameter correction schemes, and a warping machine adjustment command sequence is generated by combining defect distribution patterns. Furthermore, a virtual fabric inspection simulation system simulates a preview of the repair effect, and dynamic coupling analysis is performed between the adjustment commands and the equipment operating status to generate preventative maintenance strategies. The textile process knowledge base is a database containing various process parameters and their corresponding relationships in fabric production, used to provide targeted loom parameter correction schemes based on defect types. The warping machine adjustment command sequence refers to a series of commands used to adjust the warping machine's operating parameters; these commands are generated based on defect distribution patterns to optimize the fabric production process. The virtual fabric inspection simulation system is a system that uses computer simulation technology to generate fabric repair effects, used to preview the appearance of the repaired fabric. Dynamic coupling analysis refers to real-time analysis of the adjustment commands and the equipment operating status to ensure the stability and reliability of equipment operation and to generate preventative maintenance strategies.

[0106] Specifically, defect type coding is a coding system used to identify different types of defects, classifying and recording detected defects. The textile process knowledge base is a comprehensive database storing various process parameters and their interrelationships during fabric production, including loom speed, tension, and temperature. Based on the defect type coding, corresponding loom parameter correction schemes can be retrieved from the knowledge base, providing optimization suggestions for specific defect types. The warping machine adjustment command sequence is a series of commands generated based on defect distribution patterns. These commands are used to adjust the warping machine's operating parameters, such as tension controller settings and speed adjustments. The virtual fabric inspection simulation system generates preview images by simulating repair effects, allowing operators to visually evaluate the repair results. Dynamic coupling analysis refers to real-time analysis of adjustment commands and the actual operating status of the equipment to ensure the stability and reliability of equipment operation. Preventive maintenance strategies are generated based on the results of dynamic coupling analysis, providing forward-looking suggestions for equipment maintenance to reduce the likelihood of equipment failure.

[0107] Preferably, the construction of the textile process knowledge base may include the following steps: First, collect data on various process parameters and their corresponding relationships during fabric production; then, organize and store this data in a database. When calling a correction scheme, the corresponding parameter correction suggestions can be retrieved from the knowledge base based on the defect type code. The generation of warping machine adjustment command sequences can be based on defect distribution patterns. For example, if defects are mainly concentrated in a certain area of ​​the fabric, commands can be generated to adjust the tension or speed of that area. The implementation of the virtual fabric inspection simulation system can use computer graphics technology to simulate the appearance of the repaired fabric based on the location and type of defects. The specific process of dynamic coupling analysis may include: First, monitor the operating status parameters of the equipment in real time; then, compare and analyze these parameters with adjustment commands; finally, generate preventive maintenance strategies based on the analysis results. For example, if abnormal fluctuations are detected in the tension controller of the equipment, maintenance commands can be generated to remind operators to check and adjust. Through these detailed steps, the fabric production process can be optimized more effectively, product quality can be improved, and stable equipment operation can be ensured.

[0108] The various embodiments of the present invention have the following beneficial effects: The present invention can improve the automation level and accuracy of textile defect detection. Through multifractal feature extraction and dynamic caching technology, image data can be processed in real time; combined with adaptive clustering analysis and multi-scale annotation methods, various defects can be accurately identified and visualized heat maps can be generated; based on the defect fragment generation technology using weaving time stamps, full-process tracking of production quality problems can be achieved; by integrating real-time data from the fabric inspection machine with the process parameter library, equipment optimization suggestions can be output synchronously, forming a closed-loop quality control system.

[0109] This invention also enables intelligent management of textile production quality. By employing an improved EfficientNet module and a deformable convolutional network, it can simultaneously handle both known and unknown types of defects; through texture reconstruction algorithms and virtual repair technology, it can simulate defect compensation effects; combined with a defect level assessment model, it can achieve quantitative analysis of defect severity; and based on parameter correction schemes generated from a textile process knowledge base, it can optimize the operating status of production equipment and prevent quality problems from occurring.

[0110] like Figure 2 As shown in some embodiments, a machine vision-based textile defect detection device includes:

[0111] Image processing module 201 is used to preprocess grayscale image data of fabric surface collected during the production process of silk textiles, extract multifractal feature parameters and store fractal feature vectors in a dynamic cache pool.

[0112] Clustering analysis module 202 is used to perform adaptive clustering analysis on multiple fractal feature vectors in the cache pool to obtain multiple defect category clusters;

[0113] Typical sample determination module 203 is used to determine the typical defect samples corresponding to each cluster in the collected fabric texture map based on the mean of the fractal features of each cluster.

[0114] The annotation generation module 204 is used to obtain defect morphology annotations by processing typical defect samples through Gaussian-Laplace filtering, and to generate a heat map sequence of defect locations and weaving time stamps based on the annotations;

[0115] The image synthesis module 205 is used to generate defect-annotated enhanced images based on the heat map sequence and the original image of the fabric warp and weft structure, and to assemble the annotated images into defect segments of the loom operation cycle according to the weaving time stamp;

[0116] The data sending module 206 is used to send defect fragments and corresponding defect type codes to the fabric inspection machine quality inspection terminal.

[0117] It is understandable that the modules described in this machine vision-based textile defect detection device are similar to those in the reference. Figure 1 The steps described correspond to those in the machine vision-based textile defect detection method. Therefore, the operations, features, and beneficial effects described above for the machine vision-based textile defect detection method also apply to the machine vision-based textile defect detection device and its modules, and will not be repeated here.

[0118] The following is for reference. Figure 3 The diagram illustrates a structural schematic of a computing device 300 suitable for implementing some embodiments of the present invention. The computing device in some embodiments of the present invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 3 The terminal device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0119] like Figure 3As shown, the computing device 300 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the computing device 300. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0120] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows computing device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 A computing device 300 with various devices is shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0121] Furthermore, the storage medium in the embodiments of this application stores program instructions capable of implementing all the above methods. These program instructions can be stored in the storage medium in the form of a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or terminal devices such as computers, servers, mobile phones, and tablets.

[0122] The above description is merely a selection of preferred embodiments of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention as described in the embodiments is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. A machine vision based textile defect detection method characterized in that, The method comprises the following steps: Preprocessing the fabric surface gray image data collected in the silk textile production process, extracting multi-fractal feature parameters and storing the fractal feature vectors in a dynamic cache pool; Adaptive clustering analysis is performed on the multiple fractal feature vectors in the cache pool to obtain multiple defect category clusters; Based on the fractal feature mean value of each cluster, the typical defect samples corresponding to each cluster are determined in the collected fabric texture atlas; The typical defect samples are processed by Gaussian-Laplacian filtering to obtain corresponding defect morphology labels, and a defect position heat map sequence and corresponding weaving time stamp are generated based on the labels; Based on the heat map sequence and the warp and weft structure original map of the fabric, a defect label enhanced image is generated, and the label image is composed into a weaving machine running period defect segment according to the weaving time stamp; The defect segment and the corresponding defect type code are sent to the quality inspection terminal of the cloth inspection machine.

2. The method of claim 1, wherein, The method for determining the typical defect sample based on the fractal feature mean value of each cluster comprises the following steps: Selecting the maximum box dimension difference feature in each cluster; Determining the fabric texture primitive feature closest to the target feature, and taking the matched primitive feature sample as the typical defect sample; The method for processing the typical defect sample to obtain the defect morphology label comprises the following steps:

3. The method of claim 2, wherein, An improved EfficientNet module is used to process each cluster sample to generate defect classification confidence; Based on the confidence and the fabric defect feature database, multi-scale labeling information is generated. The method for processing the typical defect sample comprises the following steps:

4. The method of claim 3, wherein, Based on the fractal feature combination, defect feature matching retrieval is performed to determine whether the known defect mode is matched; When matching, a lightweight Ghost convolution module is used to process the sample to generate a defect type probability distribution; When not matching, a deformable convolution v2 network is used to process the sample to output an abnormal defect classification result.

5. The method according to any one of claims 1 to 4, characterized in that, The preprocessing comprises the following steps: Obtaining industrial camera collection parameters and fabric organization specification information; Performing histogram equalization processing and LOG filter enhancement on the image data to generate multi-scale texture feature maps; The method for determining the typical defect sample comprises the following steps:

6. The method according to any one of claims 1 to 4, characterized in that, Locating the yarn breakage area and the weaving density abnormal area in the multi-scale feature map. Before sending, the method further comprises the following steps: Generating a weaving machine state diagnosis report according to real-time detection data of the cloth inspection machine; 7. The method of claim 6, wherein, Spacetime alignment and fusion of the diagnosis report and the defect heat map to generate a production quality correlation atlas; Generating an equipment tuning suggestion matrix based on a textile process parameter library; Sending a full-process detection report containing the quality atlas and the tuning matrix to the terminal.

8. The method of claim 6, wherein, The label image comprises the following steps: Extracting the warp and weft yarn distribution matrix of the fabric organization cycle unit; Combining the defect position matrix and the yarn density feature to generate a weaving defect compensation matrix; Performing virtual repair rendering on the damaged yarn pixels through a texture reconstruction algorithm. The method further comprises the following steps:

9. A textile defect detection apparatus, characterized in that, Establishing a defect level evaluation model according to the fabric specification standard; When generating the compensation matrix, the defect level weight factor is fused to realize defect severity visualization labeling. The method further comprises the following steps: Calling a textile process knowledge base to generate a weaving machine parameter correction scheme based on the defect type code; Generating a warping machine adjustment instruction sequence based on the defect distribution law; Simulating a repair effect preview image through a virtual cloth inspection simulation system; Generating a preventive maintenance strategy by dynamically coupling the adjustment instruction with the equipment running state. The method comprises the following steps: The image processing module is used to preprocess grayscale image data of the fabric surface collected during the production process of silk textiles, extract multifractal feature parameters, and store the fractal feature vectors in a dynamic cache pool. The clustering analysis module is used to perform adaptive clustering analysis on multiple fractal feature vectors in the cache pool to obtain multiple defect category clusters. The typical sample determination module is used to determine the typical defect samples corresponding to each cluster in the collected fabric texture map based on the mean of the fractal features of each cluster. The annotation generation module is used to process typical defect samples through Gaussian-Laplace filtering to obtain defect morphology annotations, and generate a heat map sequence of defect locations and weaving time stamps based on the annotations; The image synthesis module is used to generate enhanced images with defect annotations based on the heat map sequence and the original image of the fabric warp and weft structure, and to assemble the annotated images into defect segments of the loom operation cycle according to the weaving time stamp; The data sending module is used to send defect fragments and corresponding defect type codes to the fabric inspection machine's quality inspection terminal.

10. A computing device, comprising: It includes a processor and a memory, the memory being used to store a computer program; when the computer program is loaded by the processor, it causes the processor to execute the machine vision-based textile defect detection method as described in any one of claims 1-8.

Images