Textile quality auditing management method and system based on big data
By combining big data and graph neural network technologies with scanning electron microscopy and graph attention networks, the problem of traditional methods failing to capture complex relationships between fibers has been solved, enabling efficient and accurate textile quality assessment and production optimization.
Patent Information
- Application Number
- CN202511155528.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-18
- Publication Date
- 2025-11-28
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional textile quality control methods struggle to capture the complex spatial relationships and diverse connections between fibers when dealing with fiber microstructure, resulting in insufficient accuracy and reliability of evaluation results, especially affecting the consistency of product performance in high-performance textiles.
A big data-based approach was adopted to acquire fiber microstructure images through scanning electron microscopy. After preprocessing, features such as fiber diameter, porosity, and surface roughness were extracted. A fiber microstructure consistency evaluation model was constructed using graph neural networks, and the importance between fibers was dynamically learned by graph attention networks (GAT) to conduct quality assessment.
It improves the precision and efficiency of textile quality control, accurately captures the complex spatial relationships between fibers, provides in-depth quality assessment, adapts to various fiber types, promptly identifies quality problems in the production process, guides production optimization, and enhances product quality and market competitiveness.
Smart Images
Figure CN121032971A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of quality inspection management technology, and in particular to a method and system for textile quality audit management based on big data. Background Technology
[0002] In the textile industry, ensuring product quality is a crucial aspect that directly impacts a product's market competitiveness. Traditional textile quality control primarily relies on manual visual inspection and some basic mechanical tests. These methods face challenges of inefficiency and insufficient accuracy when handling large-scale production. The performance of textiles is not only affected by the type of raw materials but also closely related to the microstructure of the fibers. Inconsistencies in the microstructure of fibers can lead to unstable performance of the final product, such as fluctuations in strength, elongation, and abrasion resistance. These issues are particularly critical in high-performance textiles, such as special textiles used in aerospace, military, and other fields.
[0003] With the development of computer vision and machine learning technologies, especially the breakthroughs in deep learning, new solutions have been provided for textile quality control. Through the automated analysis of images of the microstructure of textiles, these technologies can reveal complex relationships and patterns at the fiber level, providing a more accurate and detailed perspective for quality assessment.
[0004] However, relying solely on traditional image processing and deep learning methods, such as convolutional neural networks, still has limitations, especially when dealing with complex spatial relationships between fibers. These methods often neglect the interactions and connections between fibers, and their impact on overall quality. Furthermore, traditional methods struggle to capture all crucial information when processing fiber images with irregular structures and diverse connections, thus affecting the accuracy and reliability of the evaluation results.
[0005] Therefore, there is an urgent need for an efficient method that can deeply understand and analyze the complex relationships of fiber microstructure in order to improve the accuracy and efficiency of textile quality control. Summary of the Invention
[0006] To achieve the above objectives, the present invention provides a method and system for textile quality audit management based on big data.
[0007] A big data-based method for textile quality auditing and management includes the following steps:
[0008] S1. Use scanning electron microscopy (SEM) to image textile fiber samples to obtain microstructure image data of the fibers.
[0009] S2, preprocesses the collected microstructure images, including denoising, contrast enhancement and edge detection;
[0010] S3 utilizes image processing technology to extract microstructural features from preprocessed images, including fiber diameter, porosity, and surface roughness, as indicators for evaluating fiber consistency.
[0011] S4. Based on the extracted microstructure features, a fiber microstructure consistency evaluation model is constructed using a deep learning algorithm to predict fiber quality and performance.
[0012] S5. The constructed fiber microstructure consistency assessment model is used to assess the microstructure consistency of new fiber samples, and the quality grade of the fiber is determined based on the assessment results.
[0013] Furthermore, S1 specifically includes:
[0014] S11: Select representative fiber samples from the target textiles, fix the fiber samples on the sample holder for SEM imaging, and ensure that the samples are stable and fully exposed to electron beam scanning.
[0015] S12: By spraying a thin metal or carbon film, the fiber sample is made conductive, as textile fibers are usually non-conductive, to avoid the accumulation of electron beams on the sample surface and the resulting charging effect.
[0016] S13: Adjust the electron beam parameters of the SEM, including accelerating voltage, beam current and working distance, according to the characteristics of the fiber sample and imaging requirements, in order to obtain the best imaging effect and resolution.
[0017] S14: Start SEM, gradually increase the magnification, scan the fiber sample layer by layer, capture the microstructure images of the fiber from different angles and focal lengths, and obtain fiber structure information;
[0018] S15: Monitors image quality in real time during the imaging process, acquires clear images, and stores them as high-resolution digital image files.
[0019] Furthermore, the extraction of microstructural features in S3 specifically includes:
[0020] S31, Feature Localization and Segmentation: An image segmentation algorithm is applied to separate the fiber structure from the background and locate the main structure of the fiber, providing target regions for feature extraction. Based on threshold segmentation, for image I, threshold segmentation is defined as:
[0021] Where I(x,y) is the gray value of the image at point (x,y), and T is the threshold, which converts the image into a binary image for subsequent processing.
[0022] S32, Fiber Diameter Measurement: Measure and calculate the fiber diameter at different locations, based on edge detection - the Sobel operator, as follows:
[0023]
[0024] Where A is the original image, G x and G y These are the gradients in the horizontal and vertical directions, respectively, and the total gradient intensity G is obtained through... Calculations show that regions with high gradient intensity identify edges. After identifying fiber edges through edge detection, the fiber diameter is measured using the following method:
[0025] Where (x1,y1) and (x2,y2) are two points on the fiber edge, and D is the maximum measured value of the fiber diameter;
[0026] S33, Porosity Calculation: By analyzing the pixel ratio of fibers and blank areas in the segmented image, the porosity of the fibers is calculated. The porosity P is expressed by the following formula: Where A pores A is the total area of the porous region. total It is the total area of the image;
[0027] S34, Surface Roughness Assessment: Using surface texture analysis methods, the texture features of the fiber surface are extracted from the image to assess the surface roughness. Surface roughness is evaluated through parameters of the gray-level co-occurrence matrix, including contrast C and entropy E, calculated as follows:
[0028] Contrast ratio C:
[0029] Entropy E:
[0030] Where P(i,j) is the probability that gray level i and gray level j appear simultaneously in the gray-level co-occurrence matrix, contrast reflects the contrast level of the image gray-level distribution, and entropy reflects the richness of image information and is used to evaluate surface roughness.
[0031] S35, Feature Data Normalization: The extracted feature data is normalized to eliminate the influence of different dimensions and scales.
[0032] Furthermore, in S4, a graph neural network is used to construct a fiber microstructure consistency evaluation model. A graph representation is constructed, converting each fiber microstructure image into a graph representation. In this graph representation, nodes represent key points of the fiber, including fiber intersections and end points. Edges represent relationships between nodes, including fiber continuity or adjacency. Node features include fiber diameter, length, and angle, while edge features include distance or connection type between nodes. A graph attention network (GAT) is used as the main architecture, employing an attention mechanism to weight the importance of nodes, thereby capturing the interactions between nodes. GAT dynamically learns the importance of nodes by assigning different weights to different nodes.
[0033] Furthermore, the graph attention network (GAT) specifically includes:
[0034] Network Structure Design and Feature Update: A network structure with multiple graph attention layers is designed. Each graph attention layer uses a self-attention mechanism to calculate a weighted combination of node features. For node i, its updated features are represented by the following calculation:
[0035] in, Let be the feature vector of node i, N(i) be the set of neighboring nodes of node i, W be the learnable weight matrix, and α be the feature vector of node i. ij σ is the attention coefficient, representing the contribution of node j to the feature update of node i, and σ is a non-linear activation function.
[0036] Readout function - aggregation function: After all graph attention layers, an aggregation function is used to integrate the features of all nodes in the graph to generate a graph-level global feature representation for subsequent quality prediction;
[0037] Model training and optimization: The model is trained using the training dataset, and the network parameters are adjusted through the backpropagation algorithm to minimize the prediction error. During the training process, the model is validated using a validation set to avoid overfitting.
[0038] Performance evaluation: Evaluate the model's performance on the test set, using evaluation metrics (such as accuracy, recall, F1 score, etc.) to measure the model's accuracy and effectiveness in the fiber microstructure consistency evaluation task;
[0039] Model Application: The trained GAT model is applied to new fiber microstructure image data. The graph-level features extracted by the model are used to predict fiber quality and performance. Based on the relationship and pattern of fiber microstructure, a consistency score is provided for each fiber sample to indicate its quality level.
[0040] Furthermore, the attention coefficient α ij The calculation is as follows:
[0041] in, is a learnable weight vector, || denotes the concatenation operation, and LeakyReLU is a non-linear activation function.
[0042] Furthermore, the aggregation function is used to extract global features from all node features of the graph for graph-level prediction, based on average pooling, and its calculation formula is as follows:
[0043] Among them, h graph It is a feature representation at the whole graph level, where V is the set of nodes in the graph, and h i ′ is the feature vector of node i after processing by the graph attention network, and |V| is the total number of nodes in the graph. Average pooling obtains the global representation of the entire graph by calculating the average of the features of all nodes, providing input for graph-level prediction tasks, namely fiber microstructure consistency assessment.
[0044] Furthermore, the backpropagation algorithm is used to adjust network parameters through the network's gradient descent process, and the specific steps are as follows:
[0045] Forward propagation: The input, namely the fiber microstructure features represented by the graph, is propagated through the network. After processing by each layer (including the graph attention layer and the aggregation function), the output, namely the prediction of fiber quality, is finally obtained.
[0046] Loss calculation: The difference between the predicted output and the true label is calculated through a loss function, including mean squared error or cross-entropy loss. The choice of loss function depends on the specific task type (regression or classification).
[0047] Backpropagation: Calculate the gradient of the loss function with respect to each network parameter, starting from the output layer and propagating backward layer by layer using the chain rule until the input layer;
[0048] Parameter Update: Gradient descent is used to update the parameters in the network, and the update rules are as follows: Where θ is the network parameter, l is the loss function, and η is the learning rate, a hyperparameter that determines the parameter update step size.
[0049] A textile quality audit and management system based on big data, used to implement the aforementioned textile quality audit and management method based on big data, includes the following modules:
[0050] Microstructure imaging collection and preprocessing module: The scanning electron microscope is used to image textile fiber samples to obtain microstructure image data. The collected microstructure images will undergo a series of preprocessing steps, including noise reduction, contrast enhancement and edge detection.
[0051] Microstructure feature extraction module: Applying image processing technology to accurately extract microstructure features from preprocessed images, including fiber diameter, porosity, and surface roughness;
[0052] Data model building module: Based on features extracted from microstructure images, a graph neural network is used to build a fiber microstructure consistency evaluation model;
[0053] Quality assessment and judgment module: Using a trained fiber microstructure consistency assessment model, the module assesses the quality of new fiber samples and converts the consistency score output by the model into a specific quality level, providing decision support for textile production and quality control.
[0054] The beneficial effects of this invention are:
[0055] This invention, by converting the fiber microstructure into a graph representation and employing a graph attention network (GAT), can more accurately capture the complex spatial relationships and detailed features between fibers. The attention mechanism of GAT enables the model to dynamically focus on the microstructural features most critical to fiber quality assessment, such as fiber crossovers, fiber diameters, lengths, and their interactions. The detailed analysis provides a deeper and more accurate quality assessment than traditional methods, revealing potential quality problems in the microstructure, thereby guiding quality control and improvement in the production process.
[0056] This invention, through learning from the microstructure images of a large number of fiber samples, constructs a model that can not only assess the quality of specific fiber types, but also adapt to multiple types of fibers and textiles. The microstructure features and relationship patterns learned from one fiber type can also be used to assess other types of fibers. This cross-type learning and generalization ability makes the model more widely applicable in diverse applications of textile quality control.
[0057] This invention utilizes the method to accurately assess the consistency of fiber microstructure, enabling timely detection of quality issues during the production process and providing a scientific basis for production optimization. By analyzing samples with low consistency scores, potential problem areas in the production process can be traced back, such as raw material selection and textile process parameter settings, thereby achieving targeted improvements, increasing production efficiency, reducing resource waste, and also helping to improve the quality of the final product and enhance the company's market competitiveness. Attached Figure Description
[0058] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0059] Figure 1 This is a schematic diagram of the quality audit management method according to an embodiment of the present invention;
[0060] Figure 2 This is a schematic diagram of the functional modules of the quality audit management system according to an embodiment of the present invention. Detailed Implementation
[0061] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.
[0062] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0063] like Figure 1 As shown, a textile quality audit management method based on big data includes the following steps:
[0064] S1. Use scanning electron microscopy (SEM) to image textile fiber samples to obtain microstructure image data of the fibers.
[0065] S2 preprocesses the collected microstructure images, including denoising, contrast enhancement, and edge detection; median filtering is used to remove noise introduced by imaging equipment, sample preparation, or environmental factors, preserving the true information of the fiber structure; histogram equalization is used to improve the local and overall contrast of the image, making the microstructure of the fiber clearer and easier to identify in subsequent feature extraction and analysis; Canny edge detection is applied to accurately identify and extract the edge information of the fiber microstructure, highlighting the shape and texture features of the fiber.
[0066] S3 utilizes image processing technology to extract microstructural features from preprocessed images, including fiber diameter, porosity, and surface roughness, as indicators for evaluating fiber consistency.
[0067] S4. Based on the extracted microstructure features, a fiber microstructure consistency evaluation model is constructed using a deep learning algorithm to predict fiber quality and performance.
[0068] S5. The constructed fiber microstructure consistency assessment model is used to assess the microstructure consistency of new fiber samples, and the quality grade of the fiber is determined based on the assessment results.
[0069] S1 specifically includes:
[0070] S11: Select representative fiber samples from the target textiles, fix the fiber samples on the sample holder for SEM imaging, and ensure that the samples are stable and fully exposed to electron beam scanning.
[0071] S12: By spraying a thin metal or carbon film, the fiber sample is made conductive, as textile fibers are usually non-conductive, to avoid the accumulation of electron beams on the sample surface and the resulting charging effect.
[0072] S13: Adjust the electron beam parameters of the SEM, including accelerating voltage, beam current and working distance, according to the characteristics of the fiber sample and imaging requirements, in order to obtain the best imaging effect and resolution.
[0073] S14: Start SEM, gradually increase the magnification, scan the fiber sample layer by layer, capture the microstructure images of the fiber from different angles and focal lengths, and obtain fiber structure information;
[0074] S15: Monitors image quality in real time during the imaging process, acquires clear images, and stores them as high-resolution digital image files.
[0075] The extraction of microstructural features in S3 specifically includes:
[0076] S31, Feature Localization and Segmentation: An image segmentation algorithm is applied to separate the fiber structure from the background and locate the main structure of the fiber, providing target regions for feature extraction. Based on threshold segmentation, for image I, threshold segmentation is defined as:
[0077] Where I(x,y) is the gray value of the image at point (x,y), and T is the threshold, which converts the image into a binary image for subsequent processing.
[0078] S32, Fiber Diameter Measurement: Measure and calculate the fiber diameter at different locations, based on edge detection - the Sobel operator, as follows:
[0079]
[0080] Where A is the original image, G x and G y These are the gradients in the horizontal and vertical directions, respectively, and the total gradient intensity G is obtained through... Calculations show that regions with high gradient intensity identify edges. After identifying fiber edges through edge detection, the fiber diameter is measured using the following method:
[0081] Where (x1,y1) and (x2,y2) are two points on the fiber edge, and D is the maximum measured value of the fiber diameter;
[0082] S33, Porosity Calculation: By analyzing the pixel ratio of fibers and blank areas in the segmented image, the porosity of the fibers is calculated. The porosity P is expressed by the following formula: Where A pores A is the total area of the porous region. total It is the total area of the image;
[0083] S34, Surface Roughness Assessment: Using surface texture analysis methods, the texture features of the fiber surface are extracted from the image to assess the surface roughness. Surface roughness is evaluated through parameters of the gray-level co-occurrence matrix, including contrast C and entropy E, calculated as follows:
[0084] Contrast ratio C:
[0085] Entropy E:
[0086] Where P(i,j) is the probability that gray level i and gray level j appear simultaneously in the gray-level co-occurrence matrix, contrast reflects the contrast level of the image gray-level distribution, and entropy reflects the richness of image information and is used to evaluate surface roughness.
[0087] S35, Feature Data Normalization: The extracted feature data is normalized to eliminate the influence of different dimensions and scales.
[0088] In S4, a graph neural network is used to construct a fiber microstructure consistency evaluation model. A graph representation is constructed, converting each fiber microstructure image into a graph representation. In this representation, nodes represent key points of the fiber, including fiber intersections and end points, while edges represent relationships between nodes, including fiber continuity or adjacency. Node features include fiber diameter, length, and angle, while edge features include distance or connection type between nodes. A graph attention network (GAT) is used as the main architecture, employing an attention mechanism to weight the importance of nodes, thereby capturing their interactions. GAT dynamically learns the importance of nodes by assigning different weights to different nodes.
[0089] Graph Attention Network (GAT) specifically includes:
[0090] Network Structure Design and Feature Update: A network structure with multiple graph attention layers is designed. Each graph attention layer uses a self-attention mechanism to calculate a weighted combination of node features. For node i, its updated features are represented by the following calculation:
[0091] in, Let be the feature vector of node i, N(i) be the set of neighboring nodes of node i, W be the learnable weight matrix, and α be the feature vector of node i. ij σ is the attention coefficient, representing the contribution of node j to the feature update of node i, and σ is a non-linear activation function.
[0092] Readout function - aggregation function: After all graph attention layers, an aggregation function is used to integrate the features of all nodes in the graph to generate a graph-level global feature representation for subsequent quality prediction;
[0093] Model training and optimization: The model is trained using the training dataset, and the network parameters are adjusted through the backpropagation algorithm to minimize the prediction error. During the training process, the model is validated using a validation set to avoid overfitting.
[0094] Performance evaluation: Evaluate the model's performance on the test set, using evaluation metrics (such as accuracy, recall, F1 score, etc.) to measure the model's accuracy and effectiveness in the fiber microstructure consistency evaluation task;
[0095] Model Application: The trained GAT model is applied to new fiber microstructure image data. The graph-level features extracted by the model are used to predict fiber quality and performance. Based on the relationship and pattern of fiber microstructure, a consistency score is provided for each fiber sample to indicate its quality level.
[0096] The detailed scheme described above, utilizing the combination of the GAT architecture and aggregation functions, effectively handles and analyzes the complex spatial relationships of fiber microstructures, extracting valuable features for accurate quality and performance evaluation. The attention mechanism allows the model to focus on the most important parts of the graph, while the aggregation function provides a global representation of the entire graph. This combination offers a powerful tool for evaluating the consistency of fiber microstructures.
[0097] Attention coefficient α ij The calculation is as follows:
[0098] in, is a learnable weight vector, || denotes the concatenation operation, and LeakyReLU is a non-linear activation function.
[0099] Aggregation functions are used to extract global features from all node features of a graph for graph-level prediction. Based on average pooling, their calculation formula is as follows:
[0100] Among them, h graph It is a feature representation at the whole graph level, where V is the set of nodes in the graph, and h i ′ is the feature vector of node i after processing by the graph attention network, and |V| is the total number of nodes in the graph. Average pooling obtains the global representation of the entire graph by calculating the average of the features of all nodes, providing input for graph-level prediction tasks, namely fiber microstructure consistency assessment.
[0101] The backpropagation algorithm is used to adjust network parameters through the gradient descent process of the network. The specific steps are as follows:
[0102] Forward propagation: The input, namely the fiber microstructure features represented by the graph, is propagated through the network. After processing by each layer (including the graph attention layer and the aggregation function), the output, namely the prediction of fiber quality, is finally obtained.
[0103] Loss calculation: The difference between the predicted output and the true label is calculated through a loss function, including mean squared error or cross-entropy loss. The choice of loss function depends on the specific task type (regression or classification).
[0104] Backpropagation: Calculate the gradient of the loss function with respect to each network parameter, starting from the output layer and propagating backward layer by layer using the chain rule until the input layer;
[0105] Parameter Update: Gradient descent is used to update the parameters in the network, and the update rules are as follows: Where θ is the network parameter, It is the loss function, and η is the learning rate, a hyperparameter that determines the step size of parameter updates.
[0106] By repeatedly performing multiple iterations of training, the network parameters are gradually adjusted so that the predicted output is close to the true label, thereby minimizing the prediction error. In the training of the fiber microstructure consistency evaluation model, this process helps to learn to capture the complex relationship between fiber structure features and quality, so as to improve the model's prediction performance.
[0107] The graph neural network-based approach learns the complex relationships and patterns in the microstructure of fibers and then provides a consistency score for each fiber sample. The following is a detailed explanation of this process:
[0108] 1. Construction of a graphical representation of fiber samples.
[0109] First, each fiber sample is converted into a graphical representation using an image of its microstructure. In this graph:
[0110] Nodes represent key feature points of the fiber, such as intersections, fiber ends, or significant regions in the fiber's microstructure; edges represent relationships between nodes, such as fiber continuity, distances or angles between nodes; each node is assigned features, which may include the fiber's diameter, length, local texture features, etc.
[0111] 2. Application of Graph Attention Network (GAT).
[0112] Using a Graph Attention Network (GAT), the model can dynamically learn and weight the importance of each node in the graph. Key steps include:
[0113] Attention Mechanism: Through a self-attention mechanism, the model learns the importance of different nodes in influencing fiber quality. For each node, the model calculates its attention coefficient with neighboring nodes, reflecting the degree to which the features of neighboring nodes contribute to the feature update of the current node.
[0114] Feature update: Based on the attention coefficient, the features of each node are updated by weighting the features of its neighboring nodes, thereby incorporating important information about the local microstructure at each node.
[0115] 3. Application of aggregate functions.
[0116] After processing through multiple GAT layers, each node's features have acquired a rich representation containing information about its neighbors. Then, these updated node features are integrated into a full-graph-level feature representation using an aggregation function. The goal is to summarize the information of all nodes in the graph to capture the overall microstructural features and consistency of the entire fiber sample.
[0117] 4. Generation of consistency scores.
[0118] The full-image-level feature representation is further processed through one or more fully connected layers, ultimately outputting a consistency score. Multiple fully connected layers are stacked, and the calculation formula for each fully connected layer is as follows: h l+1 =σ(W l h l +b l ), where h l It is the input feature vector of the l-th layer. For the first fully connected layer, h0 = h graph W l and b l These are the weight matrix and bias term of the l-th layer, respectively; σ is a non-linear activation function, such as ReLU(σ(x)=max(0,x)); h l+1 The first layer is the output feature vector after the activation function, used as the input to the next layer. The last fully connected layer is designed as the output layer, its purpose being to transform the feature vector from the previous layer into the final consistency score. The calculation formula for the output layer is similar to that of a regular fully connected layer, but it does not contain a non-linear activation function, or it uses an activation function suitable for a specific task (such as regression or classification), for example:
[0119] For regression tasks (where the consistency score is a continuous value), the output layer is directly a linear transformation: score = W. o h o +b o For classification tasks (classifying fiber quality into several levels), the output layer uses the softmax activation function to output the probability of each class. After processing by the output layer, the final output score represents the model's evaluation of the consistency of the microstructure of the fiber sample. This score can be directly used to represent the fiber quality level.
[0120] High score: indicates that the fiber microstructure exhibits good overall consistency, suggesting high fiber quality.
[0121] Low score: This indicates that the fiber microstructure exhibits poor overall consistency, which may indicate quality problems.
[0122] 5. Model training and evaluation.
[0123] By comparing the results with actual quality labels, a loss function is calculated, and the network parameters are updated using the backpropagation algorithm to minimize the prediction error. After multiple iterations of training, the model is able to provide more accurate consistency scores for fiber samples based on the complex relationships and patterns of the fiber microstructure.
[0124] like Figure 2 As shown, a textile quality audit and management system based on big data, used to implement the aforementioned textile quality audit and management method based on big data, includes the following modules:
[0125] Microstructure imaging collection and preprocessing module: The scanning electron microscope is used to image textile fiber samples to obtain microstructure image data. The collected microstructure images will undergo a series of preprocessing steps, including noise reduction, contrast enhancement and edge detection.
[0126] Microstructure feature extraction module: Applying image processing technology to accurately extract microstructure features from preprocessed images, including fiber diameter, porosity, and surface roughness;
[0127] Data model building module: Based on features extracted from microstructure images, a graph neural network is used to build a fiber microstructure consistency evaluation model;
[0128] Quality assessment and judgment module: Using a trained fiber microstructure consistency assessment model, the module assesses the quality of new fiber samples and converts the consistency score output by the model into a specific quality level, providing decision support for textile production and quality control.
[0129] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of the invention as described above, which are not provided in detail for the sake of brevity.
[0130] This invention is intended to cover all such substitutions, modifications, and variations falling within the broad scope of the claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should be included within the scope of protection of this invention.
Claims
1. A textile quality audit and management method based on big data, characterized in that, Includes the following steps: S1. Use scanning electron microscopy (SEM) to image textile fiber samples to obtain microstructure image data of the fibers. S2, preprocesses the collected microstructure images, including denoising, contrast enhancement and edge detection; S3 utilizes image processing technology to extract microstructural features from preprocessed images, including fiber diameter, porosity, and surface roughness, as indicators for evaluating fiber consistency. S4. Based on the extracted microstructure features, a fiber microstructure consistency evaluation model is constructed using a deep learning algorithm to predict fiber quality and performance. S5. The constructed fiber microstructure consistency assessment model is used to assess the microstructure consistency of new fiber samples, and the quality grade of the fiber is determined based on the assessment results.
2. The textile quality audit and management method based on big data according to claim 1, characterized in that, S1 specifically includes: S11: Select representative fiber samples from the target textiles, fix the fiber samples on the sample holder for SEM imaging, and ensure that the samples are stable and fully exposed to electron beam scanning. S12: Conductive treatment of fiber samples by spraying thin metal or carbon film to avoid the accumulation of electron beam on the sample surface and thus prevent charging effect. S13: Adjust the electron beam parameters of the SEM, including accelerating voltage, beam current and working distance, according to the characteristics of the fiber sample and imaging requirements, in order to obtain the best imaging effect and resolution. S14: Start SEM, gradually increase the magnification, scan the fiber sample layer by layer, capture the microstructure images of the fiber from different angles and focal lengths, and obtain fiber structure information; S15: Monitors image quality in real time during the imaging process, acquires clear images, and stores them as high-resolution digital image files.
3. The textile quality audit and management method based on big data according to claim 2, characterized in that, The extraction of microstructural features in S3 specifically includes: S31, Feature Localization and Segmentation: An image segmentation algorithm is applied to separate the fiber structure from the background and locate the main structure of the fiber, providing target regions for feature extraction. Based on threshold segmentation, for image I, threshold segmentation is defined as: Where I(x,y) is the gray value of the image at point (x,y), and T is the threshold, which converts the image into a binary image for subsequent processing. S32, Fiber Diameter Measurement: Measure and calculate the fiber diameter at different locations, based on edge detection - the Sobel operator, as follows: Where A is the original image, G x and G y These are the gradients in the horizontal and vertical directions, respectively, and the total gradient intensity G is obtained through... Calculations show that regions with high gradient intensity identify edges. After identifying fiber edges through edge detection, the fiber diameter is measured using the following method: Where (x1,y1) and (x2,y2) are two points on the fiber edge, and D is the maximum measured value of the fiber diameter; S33, Porosity Calculation: By analyzing the pixel ratio of fibers and blank areas in the segmented image, the porosity of the fibers is calculated. The porosity P is expressed by the following formula: Where A pores A is the total area of the porous region. total It is the total area of the image; S34, Surface Roughness Assessment: Using surface texture analysis methods, the texture features of the fiber surface are extracted from the image to assess the surface roughness. Surface roughness is evaluated through parameters of the gray-level co-occurrence matrix, including contrast C and entropy E, calculated as follows: Contrast ratio C: Entropy E: Where P(i,j) is the probability that gray level i and gray level j appear simultaneously in the gray-level co-occurrence matrix, contrast reflects the contrast level of the image gray-level distribution, and entropy reflects the richness of image information and is used to evaluate surface roughness. S35, Feature Data Normalization: The extracted feature data is normalized to eliminate the influence of different dimensions and scales.
4. The textile quality audit and management method based on big data according to claim 1, characterized in that, In S4, a graph neural network is used to construct a fiber microstructure consistency evaluation model. A graph representation is constructed, converting each fiber microstructure image into a graph representation. In this graph representation, nodes represent key points of the fiber, including fiber intersections and end points, and edges represent relationships between nodes, including fiber continuity or adjacency. Node features include fiber diameter, length, and angle, while edge features include distance or connection type between nodes. A graph attention network (GAT) is used as the main architecture, utilizing an attention mechanism to weight the importance between nodes, thereby capturing the interactions between nodes. GAT dynamically learns the importance between nodes by assigning different weights to different nodes.
5. The textile quality audit and management method based on big data according to claim 4, characterized in that, The graph attention network (GAT) specifically includes: Network Structure Design and Feature Update: A network structure with multiple graph attention layers is designed. Each graph attention layer uses a self-attention mechanism to calculate a weighted combination of node features. For node i, its updated features are represented by the following calculation: in, Let be the feature vector of node i, N(i) be the set of neighboring nodes of node i, W be the learnable weight matrix, and α be the feature vector of node i. ij σ is the attention coefficient, representing the contribution of node j to the feature update of node i, and σ is a non-linear activation function. Readout function - aggregation function: After all graph attention layers, an aggregation function is used to integrate the features of all nodes in the graph to generate a graph-level global feature representation for subsequent quality prediction; Model training and optimization: The model is trained using the training dataset, and the network parameters are adjusted through the backpropagation algorithm to minimize the prediction error. During the training process, the model is validated using a validation set to avoid overfitting. Performance evaluation: The model's performance is evaluated on the test set, and evaluation metrics are used to measure the model's accuracy and effectiveness in the fiber microstructure consistency evaluation task; Model Application: The trained GAT model is applied to new fiber microstructure image data. The graph-level features extracted by the model are used to predict fiber quality and performance. Based on the relationship and pattern of fiber microstructure, a consistency score is provided for each fiber sample to indicate its quality level.
6. The textile quality audit and management method based on big data according to claim 5, characterized in that, The attention coefficient α ij The calculation is as follows: in, is a learnable weight vector, || denotes the concatenation operation, and LeakyReLU is a non-linear activation function.
7. The textile quality audit and management method based on big data according to claim 5, characterized in that, The aggregation function is used to extract global features from all node features of the graph for graph-level prediction. Based on average pooling, its calculation formula is as follows: Among them, h graph It is a feature representation at the whole graph level, where V is the set of nodes in the graph, and h i ′ is the feature vector of node i after processing by the graph attention network, and |V| is the total number of nodes in the graph. Average pooling obtains the global representation of the entire graph by calculating the average of the features of all nodes, providing input for graph-level prediction tasks, namely fiber microstructure consistency assessment.
8. The textile quality audit and management method based on big data according to claim 5, characterized in that, The backpropagation algorithm is used to adjust network parameters through the gradient descent process of the network. The specific steps are as follows: Forward propagation: The input, namely the fiber microstructure features represented by the graph, is propagated through the network. After processing at each layer, the output, namely the prediction of fiber quality, is finally obtained. Loss calculation: The difference between the predicted output and the true label is calculated using a loss function, including mean squared error or cross-entropy loss. Backpropagation: Calculate the gradient of the loss function with respect to each network parameter, starting from the output layer and propagating backward layer by layer using the chain rule until the input layer; Parameter Update: Gradient descent is used to update the parameters in the network, and the update rules are as follows: Where θ is the network parameter, l is the loss function, and η is the learning rate, a hyperparameter that determines the parameter update step size.
9. A textile quality audit and management system based on big data, used to implement the textile quality audit and management method based on big data as described in any one of claims 1-8, characterized in that, Includes the following modules: Microstructure imaging collection and preprocessing module: The scanning electron microscope is used to image textile fiber samples to obtain microstructure image data. The collected microstructure images will undergo a series of preprocessing steps, including noise reduction, contrast enhancement and edge detection. Microstructure feature extraction module: Applying image processing technology to accurately extract microstructure features from preprocessed images, including fiber diameter, porosity, and surface roughness; Data model building module: Based on features extracted from microstructure images, a graph neural network is used to build a fiber microstructure consistency evaluation model; Quality assessment and judgment module: Using a trained fiber microstructure consistency assessment model, the module assesses the quality of new fiber samples and converts the consistency score output by the model into a specific quality level, providing decision support for textile production and quality control.