CN118583802B - Intelligent cloth inspection detection method, system, equipment and medium - Google Patents

Abstract

The present application relates to an intelligent cloth inspection method, system, equipment and medium, belonging to the field of textile inspection technology, and the inspection method includes: obtaining physical property data and spectral response data of the cloth to be inspected, and determining the material type of the cloth to be inspected; adaptively adjusting multispectral image acquisition parameters according to the material type, performing multispectral irradiation on the cloth to be inspected and collecting image data under different spectral bands to obtain multispectral image data; performing pixel-by-pixel analysis on the pre-processed multispectral image data to identify normal areas and defective areas; marking defective areas and continuously photographing the defective areas to obtain time series images of the defective areas; obtaining the moving trajectory and change trend of the defective areas on the cloth to be inspected, obtaining the dynamic characteristics of the defective areas and determining the corresponding defect types, obtaining defect detection results and sending them to the management terminal. The present application can improve the accuracy and reliability of cloth detection.

Description

Intelligent fabric inspection method, system, device and medium

Technical Field

The present application relates to the technical field of textile inspection, and in particular to an intelligent cloth inspection method, system, device and medium.

Background Art

In the modern textile industry, fabric testing and quality control play a vital role. They are the core links to ensure that the quality of textiles is stable and meets standards at all stages from raw materials to finished products. Fabric, as an indispensable basic element in many fields such as clothing, home textiles, medical supplies and industrial materials, its quality is directly related to the comfort, durability, safety and even market competitiveness of the product.

With the continuous improvement of consumers' requirements for textile quality and the urgent need for production efficiency, automated and intelligent testing technology has become an inevitable trend for the industry's upgrading and transformation. Accurate and efficient testing methods can not only timely detect defects in the production process, such as color difference, weaving defects, stains, breaks, etc., but also effectively prevent unqualified products from entering the market, reduce economic losses, and maintain brand image.

At present, the diversity of fabric types has brought great challenges to the quality control of textiles. From natural fibers (such as cotton, linen, silk) to synthetic fibers (such as polyester and nylon), different materials have different physical characteristics and chemical properties, which put forward different requirements for detection technology and standards. Especially for high-end fabrics with complex patterns and fine textures, common intelligent fabric inspection systems often lack the ability to adapt to specific material characteristics, resulting in limitations in distinguishing defects from pattern details. At the same time, for some minor defects, due to the inability to determine the dynamic characteristics of the defects, missed detection or false alarms often occur, thus affecting the accuracy and reliability of detection.

Summary of the invention

In order to improve the accuracy and reliability of cloth detection, the present application provides an intelligent cloth inspection method, system, device and medium.

In the first aspect, the present application provides an intelligent cloth inspection method, which adopts the following technical solution:

An intelligent cloth inspection method, the inspection method comprising:

Obtaining physical property data and spectral response data of the fabric to be tested;

Determining the material type of the fabric to be detected according to the physical property data and the spectral response data;

Adaptively adjusting multispectral image acquisition parameters according to the material type;

Based on the multispectral image acquisition parameters, multispectral irradiation is performed on the fabric to be inspected and image data in different spectral bands are collected to obtain multispectral image data;

Performing image preprocessing on the multispectral image data;

Perform pixel-by-pixel analysis on the preprocessed multispectral image data according to preset logic rules and preset feature thresholds to identify normal areas and defective areas of the multispectral image data;

Marking the defective area and continuously photographing the defective area to obtain a time series image of the defective area;

Analyze the time series images to obtain the moving trajectory and change trend of the defective area on the fabric to be detected, and obtain the dynamic characteristics of the defective area;

The corresponding defect type is determined according to the dynamic characteristics of the defect area, and the defect detection result is obtained and sent to the management terminal.

By adopting the above-mentioned technical solutions and comprehensively using physical property analysis, multispectral imaging technology, high-precision image processing algorithms and time series analysis, it can not only meet the detection needs of various fabric materials, but also accurately identify various defects. At the same time, it can effectively track and classify minor defects, thereby improving the accuracy and reliability of fabric detection, greatly enhancing the quality control capabilities in the production process, providing strong technical support for the textile industry, and promoting the intelligent and sustainable development of textile inspection.

Optionally, the step of adaptively adjusting the multispectral image acquisition parameters according to the material type includes:

Acquire current environmental data in real time and perform data preprocessing;

Extracting environmental features corresponding to the preprocessed current environmental data;

constructing an input feature set based on the category features corresponding to the material type and the environmental features;

Inputting the input feature set into a pre-trained parameter configuration model, predicting optimal multispectral image acquisition parameters, and obtaining model prediction results;

According to the prediction results of the model, the multispectral image acquisition parameters are adaptively adjusted.

By adopting the above technical solutions, the image acquisition strategy is dynamically adjusted according to different material types and changing environmental conditions to ensure that the best imaging conditions can be obtained in various complex environments, so as to reduce the negative impact of external environmental factors on image quality and make defect identification more accurate. In addition, by reducing the reliance on manual intervention configuration parameters, the entire inspection process is made smoother and more efficient, thereby improving the overall inspection efficiency.

Optionally, the step of performing pixel-by-pixel analysis on the preprocessed multispectral image data according to preset logical rules and preset feature thresholds includes:

Perform image segmentation based on the preprocessed multispectral image data to obtain a plurality of homogeneous regions that are initially divided;

Extract the spectral features of each pixel in each spectral band in multiple homogeneous areas to obtain the single pixel features of each pixel;

Calculate the difference in spectral characteristics between each pixel and its adjacent pixels to obtain the neighborhood characteristics of each pixel;

According to the single pixel features and neighborhood features, a comprehensive feature vector of each pixel is obtained by weighted fusion;

Based on the preset logical rules, determine whether the comprehensive feature vector of each pixel exceeds the preset feature threshold; if so, mark the pixel as a defective pixel; if not, mark the pixel as a normal pixel;

Obtaining a defective pixel set according to the plurality of defective pixel points;

Based on an image processing algorithm, defining a defect boundary of a defect area according to the defect pixel set;

The multispectral image data is divided into a normal area and a defect area according to the defect boundary.

By adopting the above technical solution, not only the accuracy and efficiency of fabric defect detection are improved, but also the recognition ability of complex defect patterns is enhanced by comprehensive analysis of pixel-level features and spatial context information, which significantly improves the intelligence level of fabric image data analysis.

Optionally, the step of analyzing the time series images to obtain the movement trajectory and change trend of the defective area on the fabric to be detected and obtaining the dynamic characteristics of the defective area includes:

Performing image registration on the time series images of the defective area;

Extract defect feature vectors based on the registered time series images and construct a time series of defect feature vectors;

Based on the dynamic time warping algorithm, the similarity of defect feature vectors between adjacent frames is compared according to the time series of defect feature vectors, and the same defects in consecutive frames are matched;

According to the same defects in consecutive frames, a continuous moving trajectory of the defect feature vector is constructed;

Extracting defect variation parameters according to the continuous moving trajectory and identifying the variation trend of the defect based on a time series analysis algorithm;

According to the continuous movement trajectory and the change trend, the dynamic characteristics of the defect area are extracted; wherein the dynamic characteristics include diffusion rate, color change and movement path.

By adopting the above technical solutions, the dynamic behavior of the defect area in the time dimension is captured, which deepens the understanding of the dynamic characteristics of defects, helps to accurately determine the defect type, and provides more comprehensive and in-depth information for textile quality control and process optimization.

Optionally, the detection method further includes a training step of a parameter configuration model, and the training step includes:

Acquire and preprocess a historical sample feature set; the historical sample feature set includes optimal multispectral image acquisition parameters for different materials under different environments and corresponding image quality indicators;

Perform feature extraction on the preprocessed historical sample feature set to obtain a basic feature vector set;

Dividing the basic feature vector set into training samples and verification samples;

Inputting the training samples into a pre-built gradient boosting tree model for initial training, optimizing model parameters, and obtaining an initial training parameter configuration model;

The parameter configuration model is verified and model parameters are corrected based on the verification sample until the loss function of the model meets a preset condition or the number of model iterations reaches a preset number, thereby obtaining the trained parameter configuration model.

By adopting the above technical solutions, after meticulous data preprocessing, well-designed feature engineering, and efficient model training and optimization process based on gradient boosting tree, a parameter configuration model that can adaptively adjust the multispectral image acquisition parameters was constructed; this model can accurately predict the optimal image acquisition settings based on material type and real-time environmental data, significantly improving the automation level and detection accuracy of fabric detection, reducing labor costs, and providing strong technical support for intelligent production in the textile industry.

Optionally, after the step of identifying the normal area and the defective area on the fabric to be inspected, the method further includes:

Determine a normal area on the fabric to be tested as a sample test area, and obtain a sample coordinate set of the sample test area;

Perform near-infrared light reflection according to the sample coordinate set to obtain near-infrared spectrum data of the sample test area;

Preprocessing the near infrared spectrum data of the sample test area;

Inputting the pre-processed near-infrared spectral data into a pre-built machine learning model to obtain a chemical residue analysis result of the sample test area; wherein the chemical residue analysis result includes the type and concentration value of the chemical residue at each sample point in the sample test area;

Based on the preset degradable substance database, the degradability level corresponding to the fabric to be tested is evaluated according to the chemical residue analysis result and sent to the management terminal.

By adopting the above technical solution, after completing the defect detection, the test links of environmental protection performance such as chemical residue and degradability are added. By adopting non-destructive near-infrared spectroscopy analysis and detection technology, chemical residue analysis and degradability test are carried out on the tested fabrics, and relevant indicator data are quickly obtained to measure the environmental protection performance of the fabrics. This provides a scientific basis for the environmentally friendly production and consumption of textiles, which is of great significance to promoting the sustainable development of the textile industry.

Optionally, after the step of identifying the normal area and the defective area on the fabric to be inspected, the method further includes:

Determine a normal area on the fabric to be tested as a sample test area, and obtain a sample coordinate set of the sample test area;

Perform near-infrared light reflection according to the sample coordinate set to obtain near-infrared spectrum data of the sample test area;

Preprocessing the near infrared spectrum data of the sample test area;

The preprocessed spectral data are processed by first-order and second-order derivatives, and potential characteristic peaks are identified based on the peak detection algorithm;

Filter the noise peaks in the potential characteristic peaks to obtain chemical characteristic peaks;

Matching the chemical characteristic peaks with a preset chemical residue spectral characteristic library to determine the type of chemical residue at each sample point;

Calculate the concentration value corresponding to the chemical residue type at each sample point based on the standard curve method;

According to the chemical residue types and concentration values of each sample point, the chemical residue analysis results are obtained;

Based on the preset degradable substance database, the degradability level corresponding to the fabric to be tested is evaluated according to the chemical residue analysis result and sent to the management terminal.

By adopting the above technical solution, after identifying the normal area on the fabric to be tested, near-infrared spectroscopy technology is used to perform non-destructive chemical residue analysis, identify and calculate the type and concentration of chemical residues at each sample point, and then evaluate the degradability level of the fabric based on the preset degradable substance database. This solution uses clear logical rules and standardized methods to perform chemical residue analysis and degradability evaluation, ensuring the accuracy and reliability of the test results, while quickly providing environmental performance indicators of the fabric, providing a scientific basis for the environmentally friendly production and consumption of textiles, and is of great significance to promoting the sustainable development of the textile industry.

In the second aspect, the present application provides an intelligent cloth inspection system, which adopts the following technical solutions:

An intelligent cloth inspection system, the inspection system comprising:

A data acquisition module, used to acquire physical property data and spectral response data of the fabric to be tested;

A material type identification module, used to determine the material type of the fabric to be detected according to the physical property data and the spectral response data;

An acquisition parameter adjustment module, used for adaptively adjusting the multispectral image acquisition parameters according to the material type;

A multispectral image data acquisition module, used for performing multispectral irradiation on the fabric to be detected and acquiring image data in different spectral bands based on the multispectral image acquisition parameters to obtain multispectral image data;

An image preprocessing module, used for performing image preprocessing on the multispectral image data;

An image analysis module, used to perform pixel-by-pixel analysis on the preprocessed multispectral image data according to preset logic rules and preset feature thresholds, and identify normal areas and defective areas of the multispectral image data;

A marking module, used for marking the defective area;

A time series image acquisition module, used for continuously photographing the defect area to obtain a time series image of the defect area;

A dynamic characteristic generation module, used for analyzing the time series images to obtain the moving track and change trend of the defective area on the fabric to be detected, and obtaining the dynamic characteristics of the defective area;

The defect detection module is used to determine the corresponding defect type according to the dynamic characteristics of the defect area, obtain the defect detection result and send it to the management terminal.

In a third aspect, the present application provides a computer device, which adopts the following technical solution:

A computer device comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method as described in the first aspect.

In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution:

A computer-readable storage medium stores a computer program that can be loaded by a processor and execute any one of the methods in the first aspect.

To summarize, the present application includes at least one of the following beneficial technical effects: comprehensive use of physical property analysis, multispectral imaging technology, high-precision image processing algorithms and time series analysis, which can not only adapt to the detection needs of various fabric materials, but also accurately identify various defects, and effectively track and classify minor defects, thereby improving the accuracy and reliability of fabric detection, greatly enhancing the quality control capabilities in the production process, providing strong technical support for the textile industry, and promoting the intelligent and sustainable development of textile inspection.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of a first process flow of an intelligent fabric inspection method according to one embodiment of the present application;

FIG2 is a schematic diagram of a second process of the intelligent fabric inspection method according to one embodiment of the present application;

FIG3 is a schematic diagram of a third flow chart of the intelligent cloth inspection method according to one embodiment of the present application;

FIG4 is a schematic diagram of a model structure of a parameter configuration model based on a gradient boosting tree according to one embodiment of the present application;

FIG5 is a schematic diagram of a fourth process flow of the intelligent cloth inspection method according to one embodiment of the present application;

FIG6 is a schematic diagram of a fifth process flow of the intelligent fabric inspection method according to one embodiment of the present application;

FIG. 7 is a sixth flow chart of the intelligent fabric inspection method according to one embodiment of the present application;

FIG. 8 is a seventh flow chart of the intelligent fabric inspection method according to one embodiment of the present application.

DETAILED DESCRIPTION

In order to make the purpose, technical solution and advantages of the present application more clear, the present application is further described in detail below in conjunction with Figures 1-8 and embodiments. It should be understood that the specific embodiments described here are only used to explain the present application and are not used to limit the present application.

At present, fabric quality control is one of the key links to ensure the final quality of the product. Traditional fabric inspection mostly relies on manual visual inspection, which is not only inefficient, but also greatly affected by the subjective experience of the inspectors, prone to misjudgment and missed inspections, especially when facing large-scale, high-speed production lines. This limitation is more significant. With the advancement of science and technology, although some automated and semi-automated fabric inspection equipment has appeared on the market, these devices often have problems such as low detection accuracy, poor adaptability, and difficulty in handling complex patterns and minor defects. Especially for the detection of new fiber materials, traditional technology is even more difficult to meet the needs.

Especially for high-end fabrics with complex patterns and fine textures, existing automatic detection systems often lack the ability to adaptively adjust to specific material characteristics, resulting in limitations in distinguishing defects from pattern details. In addition, when detecting tiny defects, these systems often miss detections or report false alarms due to the lack of effective dynamic tracking and time series analysis technology, which affects the reliability and efficiency of detection. At the same time, the market's requirements for the environmental performance of textiles are increasing, but the integrated application of existing technologies in chemical residue analysis and degradability assessment is still insufficient, making it difficult to provide comprehensive environmental assessment reports, which restricts the sustainable development of the textile industry.

Therefore, this application aims to solve the above-mentioned technical problems and proposes an intelligent fabric inspection method that integrates multi-spectral image analysis, intelligent logical judgment, dynamic tracking and environmental assessment, aiming to achieve efficient, accurate and comprehensive control of fabric quality.

The embodiment of the present application discloses an intelligent cloth inspection method.

Referring to FIG. 1 , an intelligent cloth inspection method is shown, the inspection method comprising:

Step S101, obtaining physical property data and spectral response data of the fabric to be tested;

Specifically, before the fabric to be tested enters the testing line, it can first be touched with a micro physical probe to collect basic physical properties (such as hardness, density, thickness, elasticity, etc.); at the same time, the spectrometer quickly scans the spectrum emitted and reflected by the fabric to obtain the spectral response curve of the material to reflect the absorption and reflection characteristics of the fabric under different wavelengths of light.

Step S102, determining the material type of the fabric to be tested according to the physical property data and the spectral response data;

As an implementation method for determining the material type, since fabrics of different material types have different physical and spectral properties, a pre-established material characteristic database can be used to quickly identify the material type by comparing the physical characteristic data and spectral response data of various known fabric types; wherein the pre-established material characteristic database should at least cover natural fibers (such as cotton, linen, silk), synthetic fibers (such as polyester, nylon) and other new materials.

In some other embodiments, in order to improve recognition efficiency and accuracy, machine learning algorithms can also be used to integrate the dual information sources of physical properties and spectral data in the material feature database. Through intelligent analysis and model training, a material recognition model is established that can automatically learn the complex associations between these features and known material types, thereby forming a deep recognition capability for fabric materials, thereby achieving high-precision classification and identification of unknown fabrics. Compared with the method that relies on static database comparison, this solution has stronger adaptability and recognition accuracy.

It is understandable that by accurately identifying the fabric material, the subsequent detection parameter adjustment can be more targeted, thereby improving the detection efficiency and accuracy.

Step S103, adaptively adjusting multispectral image acquisition parameters according to the material type;

Among them, multispectral image acquisition parameters include the intensity and wavelength combination of the light source (ultraviolet light, visible light, infrared light), and the specific parameters of the image acquisition device (exposure time, resolution, etc.);

It can be understood that the system adaptively adjusts the multispectral image acquisition parameters to maximize the highlighting of possible defect characteristics in the material, ensuring that high-quality multispectral images can be obtained on different materials, providing key visual information for subsequent defect identification.

Step S104, based on the multispectral image acquisition parameters, multispectral irradiation is performed on the fabric to be inspected and image data in different spectral bands are collected to obtain multispectral image data;

In some specific embodiments, the multispectral image acquisition system applies the adjusted multispectral image acquisition parameters, integrates multiple light sources such as ultraviolet rays, infrared rays and visible light to perform multi-band irradiation on the fabric to be inspected, and synchronously captures image data through a high-sensitivity image acquisition device to form a multispectral image matrix as the acquired multispectral image data;

It can be understood that by covering multiple bands from ultraviolet to infrared, it is easy to capture defect characteristics that may be hidden under different spectra, which not only increases the depth and breadth of detection, but also can discover details that are difficult to detect with the naked eye or a single band.

Step S105, performing image preprocessing on the multispectral image data;

In some specific embodiments, the preprocessing step includes extracting key features from the multispectral image data, such as texture contrast, color uniformity, brightness difference, etc., and performing preprocessing operations such as denoising, enhancement, and contrast adjustment to improve image quality, ensure the effectiveness and accuracy of the features, and eliminate unnecessary interference factors.

Step S106, performing pixel-by-pixel analysis on the preprocessed multispectral image data according to preset logic rules and preset feature thresholds to identify normal areas and defective areas of the multispectral image data;

Among them, the pre-processed image data is subjected to pixel-level feature analysis, and a series of logical rules and feature thresholds are set based on the textile expert knowledge base. For example, foreign matter or color difference is identified by comparing the spectral differences of adjacent pixels, and it is determined whether the abnormal color change threshold is exceeded, thereby achieving refined defect positioning. Even very small defects can be identified, effectively distinguishing between normal textile structures and defective areas.

Step S107, marking the defective area and continuously photographing the defective area to obtain a time series image of the defective area;

Specifically, after preliminary analysis, the identified defective areas are located and marked, and a high-resolution camera can be used to continuously shoot at continuous time intervals to obtain time series images to obtain changes in the defective areas over the time series, which helps to distinguish between real defects and illusions caused by shooting angles, light changes, etc.

It is understandable that when faced with some minor defects, they may not be easy to detect in a single shot, but through continuous shooting and time series analysis, the cumulative effect of these defects over time can be observed, thereby improving the sensitivity of the detection system and reducing the risk of missed detection.

Step S108, analyzing the time series images to obtain the moving trajectory and change trend of the defective area on the fabric to be detected, and obtaining the dynamic characteristics of the defective area;

Among them, dynamic characteristics include diffusion rate, color change and movement path;

In some specific embodiments, by using continuous time series, the movement trajectory of defects on the fabric can be accurately tracked through image registration and dynamic tracking algorithms, and its changing trend over time can be analyzed to further extract the dynamic characteristics of the defect area, provide dynamic information for defect type judgment, and increase the detection depth.

Step S109, determining the corresponding defect type according to the dynamic characteristics of the defect area, obtaining the defect detection result and sending it to the management terminal.

Among them, after the defect type is determined, the defect detection results can be formed and fed back to the management terminal of the manager, providing specific guidance for subsequent fabric quality control.

In the above implementation, the comprehensive use of physical property analysis, multispectral imaging technology, high-precision image processing algorithms and time series analysis can not only meet the detection needs of various fabric materials, but also accurately identify various defects. At the same time, it can effectively track and classify minor defects, thereby improving the accuracy and reliability of fabric detection, greatly enhancing the quality control capabilities in the production process, providing strong technical support for the textile industry, and promoting the intelligent and sustainable development of textile inspection.

In some embodiments of the present application, defect types include diffusion defects, linear movement defects, color change defects, pinhole defects, fabric damage defects, color spot defects, etc.; specifically, diffusion defects include diffuse oil stains, liquid penetration and other defects, linear movement defects include thread shedding, scratches caused by mechanical parts and other defects, color change defects include uneven dyeing, chemical reactions and other defects, pinhole defects include defects such as needles piercing the cloth, uneven tension during weaving, etc., fabric damage defects include defects such as mechanical parts causing cloth damage, cloth being pulled and torn, and color spot defects include uneven dyeing, paint splashing, etc., which cause color spots.

As an implementation method for determining the corresponding defect type, a machine learning algorithm can be introduced. For example, an SVM (support vector machine) model can be used to extract dynamic characteristics such as diffusion rate, color change and movement path, construct a rich feature vector, and input the feature vector into the support vector machine (SVM) model. The model can use these feature vectors to map the input dynamic characteristics to the corresponding defect type through hyperplane classification in high-dimensional space, and output the judgment result, such as "diffusion type defect" or "color change type defect".

Specifically, the SVM model collects a large amount of time series image data of known defect types, extracts the corresponding dynamic characteristics, and constructs a training data set. Each training sample includes a feature vector and a corresponding defect type label. By using these training data to adjust the internal parameters of the model, the hyperplane that can best distinguish different defect types is found to achieve the mapping between dynamic characteristics and defect types. Through repeated training and verification, the model optimizes its classification ability and can eventually accurately determine the defect type of unknown samples in actual detection.

As another implementation method for determining the corresponding defect type, a series of logical rules based on expert knowledge and experience may be set to analyze the dynamic characteristics to determine whether they conform to certain specific defect patterns, so as to determine the defect type.

Specifically, for example, if the dynamic characteristics show that the area of the defect area spreads rapidly over time and is accompanied by color changes, it can be determined as a "diffusion type defect (such as oil diffusion)"; if the position change shows regular movement, it can be determined as a "linear movement type defect (such as thread loss)"; if the spectral characteristics change significantly, it can be determined as a "color change type defect (such as uneven dyeing)". In this way, the system can accurately identify and classify different types of defects, such as "diffusion type oil stains", "linear moving fabric defects" or "gradually changing color spots".

2 , as an implementation of step S103 , the step of adaptively adjusting the multispectral image acquisition parameters according to the material type includes:

Step S201, real-time acquisition of current environment data and data preprocessing;

In some specific embodiments, the surrounding environment state can be monitored in real time through a sensor network (such as light sensors, temperature and humidity sensors) to collect key parameters such as light intensity, temperature, and humidity. The data preprocessing step includes data cleaning and standardization. By removing outliers, the numerical data such as light intensity and temperature are normalized to ensure data quality and facilitate subsequent analysis.

Step S202, extracting environmental features corresponding to the preprocessed current environmental data;

Specifically, from the pre-processed environmental data, feature variables closely related to the image acquisition quality are selected, such as normalized light intensity (affecting exposure time and light source compensation), temperature (which may affect camera component performance), and other features, to directly or indirectly reflect the impact of the current environment on the imaging quality;

It can be understood that by focusing on the key factors affecting the acquisition parameters, the pertinence and accuracy of the model prediction can be improved.

Step S203, constructing an input feature set based on the category features and environmental features corresponding to the material type;

Among them, a comprehensive input feature set is constructed by combining the identified material type (as a category feature, reflecting the differences in spectral absorption and reflection of different materials) and the extracted environmental features, aiming to form a feature vector set that can effectively express the complexity of the acquisition scene.

Step S204, inputting the input feature set into a pre-trained parameter configuration model, predicting the optimal multispectral image acquisition parameters, and obtaining a model prediction result;

Among them, the parameter configuration model learns the optimal configuration mapping relationship between material properties and environmental factors to acquisition parameters (such as light source intensity, wavelength combination, exposure time, etc.) based on historical data and machine learning algorithms. Therefore, it can quickly and accurately predict the optimal acquisition parameters under current specific conditions, and realize intelligent and adaptive parameter configuration.

Step S205: adaptively adjust the multispectral image acquisition parameters according to the model prediction results.

Among them, according to the model prediction results, the system automatically adjusts the parameter configuration of the acquisition equipment, such as adjusting the brightness of the light source, switching to the recommended spectral band combination, and setting the camera's exposure parameters and resolution, etc., in order to obtain the highest quality image data.

In the above implementation, the image acquisition strategy is dynamically adjusted according to different material types and changing environmental conditions to ensure that optimal imaging conditions can be obtained in various complex environments, so as to reduce the negative impact of external environmental factors on image quality and make defect identification more accurate; in addition, by reducing the reliance on manual intervention configuration parameters, the entire detection process is made smoother and more efficient, thereby improving the overall detection efficiency.

It is understandable that adaptive adjustment generally refers to the system automatically adjusting its parameters or behaviors according to environmental changes or conditions to better adapt to the current situation or needs; therefore, the present application introduces real-time environmental data in the process of optimizing and adjusting multispectral image acquisition parameters, and ensures the best image quality under various conditions by realizing dynamic adaptive adjustment of acquisition parameters, thereby improving the stability of the detection system and the reliability of the detection results.

For example, when collecting multispectral images of a certain type of cotton fabric with strong hygroscopicity in a high humidity environment, if environmental factors are not taken into account, high humidity may cause the micro-reflection on the surface of the fabric to increase, affecting the contrast and clarity of the image; the introduction of real-time environmental data enables the system to identify the current high humidity conditions, automatically adjust the light source intensity and select a wavelength combination that is more conducive to reducing the impact of surface reflections, while adjusting the camera's exposure time to reduce image blur caused by ambient humidity, ensure image quality, and accurately detect fabric defects.

For example, when the light intensity changes, the system can automatically adjust the camera's exposure time and gain to maintain appropriate brightness and contrast of the image; when the temperature changes, the image quality and accuracy of the spectral data can be stabilized by adjusting the light source power or compensation algorithm.

3 , as a further implementation of the intelligent fabric inspection method, the inspection method further includes a training step of a parameter configuration model, and the training step includes:

Step S301, obtaining a historical sample feature set and performing preprocessing;

Among them, the historical sample feature set includes the optimal multispectral image acquisition parameters of different materials in different environments and the corresponding image quality indicators;

Specifically, detailed parameter records of multispectral images that have been successfully collected in the past can be collected, including the optimal acquisition parameters such as light source intensity, wavelength combination, exposure time, etc. used for different materials in specific environments (such as different lighting and temperature conditions), as well as image quality indicators (such as clarity and contrast) obtained under these parameters.

In some embodiments, the preprocessing step includes data cleaning, standardization and normalization to ensure data quality and efficiency of model training; by constructing a rich training data set and eliminating noise and inconsistency in the data through data preprocessing, a solid foundation is laid for model training.

Step S302, extracting features from the preprocessed historical sample feature set to obtain a basic feature vector set;

Among them, features closely related to image acquisition quality are extracted from the preprocessed data, such as material type, environmental factors (light intensity, temperature) and acquisition parameters (light source intensity, etc.). These features are combined and transformed to form a basic feature vector set to provide input for the model; through carefully designed feature engineering, the model's sensitivity to key factors is improved, which helps the model better understand and learn patterns in the data.

Step S303, dividing the basic feature vector set into training samples and verification samples;

Among them, the basic feature vector set is randomly divided into two parts, one part is used for model training, and the other part is used as a validation set to evaluate the model performance and ensure the generalization ability of the model.

Step S304, inputting the training samples into a pre-built gradient boosting tree model for initial training, optimizing the model parameters, and obtaining an initial training parameter configuration model;

In an embodiment of the present application, a gradient boosted tree (GBM) model may be used for initial training. GBM gradually improves prediction performance by iteratively adding weak learners (usually decision trees), and each tree attempts to correct the residual of the previous tree. The initial training stage includes optimizing the hyperparameters of the model, such as the learning rate, the number of trees, the maximum depth of the trees, etc. Through the iterative learning of GBM, the model can gradually learn the nonlinear relationship between complex features and gradually improve the prediction accuracy.

Step S305, verifying the parameter configuration model and correcting the model parameters based on the verification sample until the loss function of the model meets the preset conditions or the number of model iterations reaches the preset number, thereby obtaining a trained parameter configuration model.

Specifically, the validation set is used to evaluate the performance of the initial training model, and the model parameters are adjusted according to the evaluation results, such as reducing the learning rate, increasing the complexity of the tree, etc., until the loss function of the model on the validation set reaches a preset threshold or reaches a predetermined number of iterations, ensuring that the model achieves the best prediction performance while maintaining good generalization ability, thereby improving the practicality and stability of the model.

In the above implementation, after meticulous data preprocessing, carefully designed feature engineering, and efficient model training and optimization process based on gradient boosting tree, a parameter configuration model that can adaptively adjust multispectral image acquisition parameters is constructed; the model can accurately predict the optimal image acquisition settings according to material type and real-time environmental data, significantly improving the automation level and detection accuracy of fabric detection, reducing labor costs, and providing strong technical support for intelligent production in the textile industry.

As a specific implementation method for constructing a parameter configuration model based on a gradient boosting tree (GBM), the GBM model gradually constructs a sequence of decision trees in an iterative manner during the training process. Each decision tree acts as a weak learner and has a fixed maximum depth (e.g., 5-10 layers) to avoid overfitting. Each internal node of the tree represents a feature test (e.g., light intensity threshold, material type classification), the branch represents the test result, and the leaf node stores the predicted acquisition parameter adjustment value (e.g., light source intensity increment, wavelength adjustment range) or regression value (e.g., exposure time). In addition, during the growth of the tree, feature importance evaluation (e.g., Gini impurity or information gain) is used to select the best split feature to ensure that each split step can minimize the prediction error.

In the specific iterative process, the GBM model gradually reduces the prediction error by adding new decision trees. Each tree is trained based on the residual of the previous tree, that is, it tries to correct the prediction error of the previous tree. The prediction output of each tree is multiplied by a learning rate less than 1 (for example, 0.1) to control the speed of model learning and reduce the risk of overfitting. In addition, the square loss or absolute loss function can be used to measure the gap between the current model prediction and the true label (optimal acquisition parameters) to guide the growth of the tree in each round of iteration.

Finally, the model is regularized by limiting the depth of the tree, splitting nodes with the minimum number of samples, and other methods to avoid excessive model complexity; when the improvement of the loss function on the validation set is less than the preset threshold (such as 0.001), or reaches the preset maximum number of iterations (such as 100 times), stop adding new decision trees and complete model training.

4 is a schematic diagram of a model structure of a parameter configuration model based on a gradient boosting tree (GBM) in one embodiment of the present application, which mainly includes the following structures:

Model starting point: Taking the "GBM model structure" as the starting point, it indicates the initialization stage of the model, ready to receive and process input data.

Feature Test Node: The model is divided into three main feature test paths, corresponding to the comprehensive evaluation of light intensity, material type and environmental factors; these feature tests are the basis of the model's decision-making process.

Light intensity decision tree: In this path, the model will make a judgment based on the light intensity data obtained in real time: if the light intensity is "lower than the threshold", the model will decide to "reduce the light intensity"; if the light intensity is "higher than the threshold", the model will decide to "increase the light intensity"; if the light intensity is "close to the ideal value", the model will maintain the "light intensity".

Material type decision tree: In the material type test, the model classifies materials according to different material properties and specifies different wavelength adjustment strategies for each material type: for "cotton material", the model will "adjust wavelength range A"; for "chemical fiber material", the model will "adjust wavelength range B"; for "blended material", the model will "adjust wavelength range C".

Comprehensive Assessment Decision Tree: The comprehensive assessment of environmental factors will be based on the output of the first two decision trees and other environmental parameters, and a comprehensive analysis will be conducted to "determine the exposure time."

Parameter output: All decision paths eventually converge to the "Output acquisition parameters" node, where the model will output the final acquisition parameters, including light source intensity, wavelength range, and exposure time, to achieve the optimal multispectral image acquisition effect.

After training is completed in the above manner, the parameter configuration model can receive real-time material type and environmental data as input, output optimized acquisition parameters, guide the adaptive adjustment of multispectral image acquisition equipment, ensure the image acquisition quality under different conditions, effectively support the intelligent fabric inspection process, improve inspection efficiency and accuracy, and meet the textile industry's needs for efficient and precise quality control.

5 , as an implementation of step S106 , the step of performing pixel-by-pixel analysis on the preprocessed multispectral image data according to preset logic rules and preset feature thresholds includes:

Step S401, performing image segmentation according to the pre-processed multispectral image data to obtain a plurality of homogeneous regions that are initially divided;

Among them, image segmentation technology can be used to preliminarily divide the multispectral image data into multiple different types of homogeneous regions based on spectral characteristics, color, texture and other information. A homogeneous region refers to a continuous area with similar attributes or characteristics, such as a set of pixels with similar color, texture, grayscale value or spectral reflectance. In some embodiments, the image segmentation technology can use threshold segmentation, edge detection or cluster analysis.

In an embodiment of the present application, these homogeneous regions may correspond to different fabric materials, color blocks or potential defective areas, and by providing a structured analysis framework for subsequent pixel-level analysis, unnecessary global searches are reduced and analysis efficiency is improved.

Step S402, extracting the spectral features of each pixel in each spectral band in multiple homogeneous regions to obtain a single pixel feature of each pixel;

The radiation intensity or reflectivity of each pixel in each spectral band is extracted to form a spectral feature vector of the pixel to reflect its spectral reflection or absorption characteristics, which is convenient for the subsequent identification of specific substances or abnormal spectral signals.

It is understandable that in the multiple homogeneous regions preliminarily divided, detailed spectral feature extraction is performed for each pixel. This means that the feature extraction of each pixel is not only based on its response in each spectral channel, but also takes into account the prior knowledge of the region where it is located (the type of homogeneous region preliminarily divided); for example, if a region is preliminarily judged as a potential defect area, the feature analysis of the pixels in the region will focus on finding abnormal spectral features.

Step S403, calculating the difference in spectral characteristics between each pixel and adjacent pixels to obtain a neighborhood feature of each pixel;

Among them, when extracting the features of a single pixel, the difference between the pixel and the surrounding adjacent pixels is considered. By calculating the statistical characteristics of the pixels in the neighborhood (such as mean, standard deviation, correlation, etc.), neighborhood features are generated to reflect the local texture and change patterns, thereby enriching the feature description of a single pixel. The feature extraction is not only based on single-point information, but also integrates local contextual information, which can further reveal local change trends and help identify subtle defects.

Step S404, obtaining a comprehensive feature vector for each pixel by weighted fusion based on single pixel features and neighborhood features;

Specifically, by combining single pixel features and neighborhood features, a comprehensive feature vector can be constructed for each pixel through weighted fusion, which can more comprehensively represent the information of the pixel and its surrounding environment, forming a comprehensive pixel feature description. It should be noted that in the weighted fusion process, the specific weight allocation rule can be based on the importance of the feature and its relevance to defect recognition.

Step S405, based on the preset logic rules, determine whether the comprehensive feature vector of each pixel exceeds the preset feature threshold. If not, jump to step S406; if yes, jump to step S407;

Step S406, marking the pixel as a normal pixel;

Step S407, marking the pixel as a defective pixel;

Among them, according to the specific needs of textile defect identification, a series of preset logical rules and feature thresholds are set to distinguish normal and abnormal spectral feature patterns.

Specifically, by evaluating the comprehensive feature vector of each pixel point, it is determined whether its feature value exceeds a preset threshold under a preset logical rule, so as to determine whether the pixel point belongs to the defect or normal category.

In the specific embodiments of the present application, as some exemplary preset logical rules and characteristic thresholds, for example: according to the spectral shift rule, if the Euclidean distance between the spectral characteristic vector of a pixel point and the average spectral characteristic vector of a typical material in the homogeneous region exceeds the characteristic threshold T1, the pixel point is determined to be a defective pixel point; wherein T1 is a threshold determined according to the spectral variation range of normal materials, that is, the characteristic threshold corresponding to the spectral shift rule. Another example: according to the neighborhood consistency rule, for any pixel point, if its neighborhood feature (such as the average or standard deviation of the spectral difference) exceeds the characteristic threshold T2, it means that the texture or color of the homogeneous region is discontinuous, and the pixel point is determined to be a defective pixel point; wherein T2 is a threshold determined by analyzing the natural variation of the spectral features of adjacent pixels in the normal region, that is, the characteristic threshold corresponding to the neighborhood consistency rule.

Step S408, obtaining a defective pixel set according to the plurality of defective pixel points;

Step S409, based on the image processing algorithm, defining the defect boundary of the defect area according to the defect pixel set;

In some embodiments, an image processing algorithm, such as a morphological operation or a cluster analysis algorithm, may be applied to define the precise boundary of the defective area based on a set of defective pixels, accurately outline the defect range, and achieve refined positioning.

Specifically, by identifying and gathering the defective pixels that are initially determined to form a set of defective pixels, the morphological dilation operation is applied to expand the influence domain of these pixels and effectively connect the adjacent defective areas. At the same time, the corrosion operation is used to remove small interference noise to ensure the accuracy of boundary definition. On this basis, clustering analysis algorithms such as DBSCAN (density-based spatial clustering application and noise filtering) are implemented to further refine the outline of the defective area and accurately outline the defective boundary based on the spectral feature similarity and spatial proximity of the pixels.

Step S410 , dividing the multispectral image data into a normal area and a defect area according to the defect boundary.

In the above implementation, not only the accuracy and efficiency of fabric defect detection are improved, but also the recognition capability of complex defect patterns is enhanced by comprehensively analyzing pixel-level features and spatial context information, thereby significantly improving the intelligence level of fabric image data analysis.

6 , as an implementation of step S108, the steps of analyzing the time series images to obtain the moving trajectory and change trend of the defective area on the fabric to be detected and obtaining the dynamic characteristics of the defective area include:

Step S501, performing image registration on the time series images of the defect area;

Among them, time series images may have inter-frame displacement due to slight changes in the shooting time (such as camera shake, cloth movement), so image registration technology is used to correct each frame of the sequence by finding common feature points between images or using transformation models (such as affine transformation, perspective transformation) to ensure that all images are aligned in the same coordinate system, thereby providing a consistent benchmark for subsequent feature extraction and matching, and reducing mismatching caused by image displacement.

Step S502, extracting defect feature vectors according to the registered time series images, and constructing a time series of defect feature vectors;

Specifically, in the registered images, high-dimensional feature vectors are extracted for the defect area in each frame of the sequence image, including but not limited to color histogram, texture features, shape descriptors, etc. These feature vectors can summarize the appearance and position information of the defects at each time point. By arranging these feature vectors in chronological order, a time series of defect features can be obtained.

Step S503, based on the dynamic time warping algorithm, the similarity of the defect feature vectors between adjacent frames is compared according to the time series of the defect feature vectors, and the same defect in consecutive frames is matched;

Among them, the dynamic time warping algorithm (DTW) is a time series comparison method that can handle the matching problem of sequences of different lengths and is particularly suitable for nonlinear alignment.

In an embodiment of the present application, for a defect feature vector sequence, by calculating and minimizing the cumulative distance cost matrix, the time scale difference is automatically adjusted to find the best matching path. Even if the defect movement speed is uneven, the same defect in consecutive frames can be effectively matched. This technical solution overcomes the limitations of strict synchronization of time series in traditional time series analysis, improves the flexibility and accuracy of matching, and ensures the continuity and integrity of the defect trajectory.

Step S504, constructing a continuous moving trajectory of the defect feature vector according to the same defect in the consecutive frames;

Among them, based on the matching results of the dynamic time warping algorithm, the feature vectors of the same defects in continuous frames are used to record the position information of the defects at each time point to construct the continuous movement trajectory of the defects on the fabric, intuitively showing the movement path of the defects.

Step S505, extracting defect change parameters according to the continuous moving trajectory and identifying the defect change trend based on a time series analysis algorithm;

Among them, by analyzing the defect movement trajectory, the defect movement speed, direction change and other parameters are calculated. At the same time, time series analysis algorithms (such as trend analysis and periodic analysis) are applied to identify the pattern of defect change, such as acceleration, deceleration, stable state or periodic change, providing a basis for understanding the defect type.

Step S506, extracting dynamic characteristics of the defect area according to the continuous movement trajectory and change trend; wherein the dynamic characteristics include diffusion rate, color change and movement path.

Specifically, the diffusion rate is the rate of change of the defect area over time, the color change is the change of the spectral characteristics of the defect area over time, and the movement path is the movement trajectory of the defect area on the cloth.

In the above implementation, the dynamic behavior of the defect area in the time dimension is captured, which deepens the understanding of the dynamic characteristics of the defect, helps to accurately determine the defect type, and provides more comprehensive and in-depth information for textile quality control and process optimization.

At present, with the increasing awareness of environmental protection, fabric testing not only focuses on the physical properties of products, but also includes the assessment of environmental impact. Chemical residues, biodegradability of dyes, energy consumption in the production process, etc. are all important indicators to measure whether textiles are environmentally friendly. Therefore, modern fabric quality control systems also need to incorporate eco-friendly assessments to ensure that the entire production chain from source to terminal meets green production standards.

7 , as a further implementation of the intelligent cloth inspection method, after the step of identifying the normal area and the defective area on the cloth to be inspected, the method further includes:

Step S601, determining a normal area on the fabric to be tested as a sample test area, and obtaining a sample coordinate set of the sample test area;

It is understandable that the normal area refers to the part without obvious physical damage, color deviation or structural abnormality. Since these areas can better represent the real physical properties and chemical composition of the fabric and avoid the interference of defects on the analysis results, the normal area is determined as the sample test area. Among them, these areas are accurately located through the coordinate system to form a sample coordinate set, which provides an accurate target position for subsequent near-infrared spectrum detection.

Step S602, performing near-infrared light reflection according to the sample coordinate set to obtain near-infrared spectrum data of the sample test area;

Among them, near-infrared spectroscopy analysis is a non-destructive testing technology that uses the absorption, transmission or reflection phenomena that occur when near-infrared light interacts with chemical components in fabrics to infer the composition and content of substances; in an embodiment of the present application, an integrated near-infrared spectrometer can be used to emit near-infrared light to the fabric area where the sample coordinates are concentrated, and receive reflected or transmitted spectral data to capture absorption peak information unique to different chemical substances.

It should be noted that the system can automatically adjust the scanning range and resolution of the spectrometer to best match the absorption peaks of common chemicals in textiles.

Step S603, preprocessing the near infrared spectrum data of the sample test area;

Among them, the original spectral data usually contains various noises and baseline drifts, which affect the analysis accuracy.

In some embodiments, the spectral data may be subjected to preprocessing steps such as baseline correction, smoothing, and noise removal to improve signal quality. For example, SNV (standard normal variable transformation) may be used to eliminate systematic variation between spectra, and MSC (multiple scatter correction) may be used to reduce background interference between spectra.

Step S604, inputting the pre-processed near infrared spectrum data into a pre-built machine learning model to obtain a chemical residue analysis result of the sample test area; wherein the chemical residue analysis result includes the type and concentration value of the chemical residue at each sample point in the sample test area;

Among them, a pre-trained machine learning model is used to take the preprocessed spectral data as input. The model matches the spectral features with the chemical residue feature patterns learned during the training process, and outputs the type of chemical residue and its approximate concentration at each sample point. This is based on the complex associations and pattern recognition capabilities learned by the model from a large number of known samples during the training phase.

In one embodiment of the present application, the machine learning model may adopt a random forest (RF) model. Specifically, during the model training process, a large number of subsets are randomly extracted from the labeled spectral data set to construct multiple decision trees, and each tree is grown based on a random subset of features to reduce the risk of overfitting and improve the generalization ability of the model. The training data is resampled by Bootstrap sampling technology. During the training phase, the model learns the complex patterns in the spectral data related to the types and concentrations of chemical residues. As each tree is established and merged, the random forest gradually optimizes the internal node splitting rules until all trees are built. The final integrated model can efficiently process new near-infrared spectral data, predict the types of chemical residues in the sample and the corresponding concentration values, thereby providing a powerful tool for the chemical composition analysis of fabrics.

Step S605: Based on the preset degradable substance database, the degradability level corresponding to the fabric to be tested is evaluated according to the chemical residue analysis result and sent to the management terminal.

Among them, the degradability level reflects the possibility of fabric decomposition in the natural environment or under specific conditions, which is crucial for evaluating its environmental friendliness and sustainability. For example, the degradability of fabric can be divided into different levels (such as high, medium, and low) according to the evaluation rules.

Specifically, based on a known database of degradable substances, the chemical residue analysis results are compared with the types of substances and degradation properties in the database, and the overall degradability potential of the fabric is evaluated based on the degradable properties of the chemical residues. For example, if the chemical residues on the fabric are mainly biodegradable substances, the fabric has a higher degradability level; otherwise, the degradability level is lower.

In the above implementation, after completing the defect detection, the test links of environmental protection performance such as chemical residue and degradability are added. By adopting non-destructive near-infrared spectroscopy analysis and detection technology, chemical residue analysis and degradability test are performed on the fabric to be tested, and relevant index data are quickly obtained to measure the environmental protection performance of the fabric, which provides a scientific basis for the environmentally friendly production and consumption of textiles, and is of great significance to promoting the sustainable development of the textile industry.

8 , as another embodiment of obtaining the degradability level corresponding to the fabric to be tested, after the step of identifying the normal area and the defective area on the fabric to be tested, the following step is further included:

Step S701, determining a normal area on the fabric to be tested as a sample test area, and obtaining a sample coordinate set of the sample test area;

The normal area is the part without defects or pollution. These areas represent the basic state of the fabric and are suitable as samples for chemical composition analysis. By determining the normal area as the sample test area, it is ensured that the collected spectral data can reflect the actual situation of the fabric.

Step S702, performing near-infrared light reflection according to the sample coordinate set to obtain near-infrared spectrum data of the sample test area;

Among them, a near-infrared light source can be used to illuminate each coordinate point of the sample test area. The near-infrared light is reflected and collected by a spectrometer. The near-infrared spectral data reflects the reflection characteristics of the sample test area at different wavelengths, and these characteristics are related to the chemical composition of the sample.

Step S703, preprocessing the near infrared spectrum data of the sample test area;

Step S704, performing first-order and second-order derivative processing on the preprocessed spectral data, and identifying potential characteristic peaks based on a peak detection algorithm;

Specifically, by performing first-order and second-order derivative processing, the significance of the characteristic peak can be enhanced, and the peak detection algorithm can adopt, for example, a Savitzky-Golay smoothing filter algorithm to identify potential characteristic peaks on the spectrum curve.

Step S705, filtering the noise peaks in the potential characteristic peaks to obtain chemical characteristic peaks;

Among them, the identified potential characteristic peaks can be filtered by setting appropriate thresholds and minimum peak heights to retain significant chemical characteristic peaks.

Step S706, matching the chemical characteristic peaks with a preset chemical residue spectral characteristic library to determine the type of chemical residue at each sample point;

Specifically, a chemical residue spectral feature library is established in advance, which contains the spectral features of various known chemical substances. These significant chemical characteristic peaks are matched with the spectral features in the chemical residue spectral feature library, and the similarity is calculated to confirm whether they are consistent with the characteristic peaks of known chemical substances. If they are consistent, the type of chemical residue at each sample point can be determined.

Step S707, calculating the concentration value corresponding to the chemical residue type of each sample point based on the standard curve method;

Among them, based on the standard curve method, the concentration of the residue at each sample point can be calculated using a calibration curve established using standard samples of known concentrations.

Step S708, obtaining chemical residue analysis results according to the chemical residue type and concentration value of each sample point;

Among them, the chemical residue types and concentration values of all sample points are summarized to generate the overall chemical residue analysis results, providing data support for subsequent evaluations.

Step S709: Based on the preset degradable substance database, the degradability level corresponding to the fabric to be tested is evaluated according to the chemical residue analysis result and sent to the management terminal.

Specifically, a database of degradable substances is established, which records the degradability information of various chemical substances; the degradability of the fabric is evaluated based on the types and concentrations of chemical residues in the chemical residue analysis results. The degradability of the fabric can be divided into different levels (such as high, medium, and low). The higher the proportion of degradable substances, the higher the degradability rating of the fabric.

In the above implementation, after identifying the normal area on the fabric to be tested, near-infrared spectroscopy is used to perform non-destructive chemical residue analysis, identify and calculate the type and concentration of chemical residues at each sample point, and then evaluate the degradability level of the fabric based on a preset database of degradable substances. This solution uses clear logical rules and standardized methods (such as the standard curve method) to perform chemical residue analysis and degradability evaluation, ensuring the accuracy and reliability of the test results, while quickly providing environmental performance indicators of the fabric, providing a scientific basis for the environmentally friendly production and consumption of textiles, and is of great significance to promoting the sustainable development of the textile industry.

For the above two specific implementation methods, in comparison, the former adopts machine learning algorithms, which have advantages in processing complex pattern recognition and large-scale data analysis, and is suitable for situations with sufficient data and abundant computing resources; while the latter adopts non-machine learning algorithms, which are highly interpretable, transparent and robust in chemical residue analysis and degradability assessment, making it more applicable in application scenarios that require stable, transparent and reliable results.

As a further implementation method of the embodiment of the present application, after obtaining the degradability level corresponding to the fabric to be tested, all test results, including defect detection results, chemical residue analysis results, degradability level and corresponding environmental indicators, can be integrated to generate a detailed fabric quality and environmental assessment report, thereby providing a basis for production decisions and rating the sustainability of the fabric to meet the textile industry's higher pursuit of environmental protection and sustainable development.

The embodiment of the present application also discloses an intelligent cloth inspection system.

An intelligent cloth inspection system, the inspection system comprising:

A data acquisition module, used to acquire physical property data and spectral response data of the fabric to be tested;

A material type identification module, used to determine the material type of the fabric to be detected based on the physical property data and the spectral response data;

An acquisition parameter adjustment module is used to adaptively adjust the multispectral image acquisition parameters according to the material type;

A multispectral image data acquisition module is used to perform multispectral irradiation on the fabric to be inspected and collect image data in different spectral bands based on multispectral image acquisition parameters to obtain multispectral image data;

An image preprocessing module, used for performing image preprocessing on multispectral image data;

An image analysis module, used to analyze the preprocessed multispectral image data pixel by pixel according to preset logic rules and preset feature thresholds, and identify normal areas and defective areas of the multispectral image data;

A marking module, used to mark defective areas;

A time series image acquisition module is used to continuously shoot the defect area to obtain a time series image of the defect area;

A dynamic characteristics generation module is used to analyze the time series images to obtain the movement trajectory and change trend of the defective area on the fabric to be detected, and obtain the dynamic characteristics of the defective area;

The defect detection module is used to determine the corresponding defect type according to the dynamic characteristics of the defect area, obtain the defect detection result and send it to the management terminal.

The intelligent fabric inspection system of the embodiment of the present application can implement any of the above-mentioned intelligent fabric inspection methods, and the specific working process of each module in the intelligent fabric inspection system can refer to the corresponding process in the above-mentioned method embodiment.

In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are only illustrative; for example, the division of a module is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

The embodiment of the present application also discloses a computer device.

The computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the intelligent fabric inspection method as described above is implemented.

The embodiment of the present application also discloses a computer-readable storage medium.

A computer-readable storage medium stores a computer program that can be loaded by a processor and execute any one of the above-mentioned intelligent fabric inspection methods.

Among them, computer-readable storage media can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus or device; the program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

It should be noted that in the above embodiments, the description of each embodiment has different emphases, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

The above are all preferred embodiments of the present application, and are not intended to limit the protection scope of the present application. Any feature disclosed in this specification (including the abstract and drawings), unless otherwise stated, can be replaced by other equivalent or alternative features with similar purposes. That is, unless otherwise stated, each feature is only an example of a series of equivalent or similar features.

Claims (8)

1. An intelligent cloth inspection method, characterized in that the inspection method comprises: obtaining physical property data and spectral response data of the cloth to be inspected; Determining the material type of the fabric to be detected according to the physical property data and the spectral response data; Adaptively adjusting multispectral image acquisition parameters according to the material type; Based on the multispectral image acquisition parameters, multispectral irradiation is performed on the fabric to be inspected and image data in different spectral bands are collected to obtain multispectral image data; Performing image preprocessing on the multispectral image data; Perform pixel-by-pixel analysis on the preprocessed multispectral image data according to preset logic rules and preset feature thresholds to identify normal areas and defective areas of the multispectral image data; Marking the defective area and continuously photographing the defective area to obtain a time series image of the defective area; Analyze the time series images to obtain the moving trajectory and change trend of the defective area on the fabric to be detected, and obtain the dynamic characteristics of the defective area; Determine the corresponding defect type according to the dynamic characteristics of the defect area, obtain the defect detection result and send it to the management terminal; According to the material type, the step of adaptively adjusting the multispectral image acquisition parameters comprises: Acquire current environmental data in real time and perform data preprocessing; Extracting environmental features corresponding to the preprocessed current environmental data; constructing an input feature set based on the category features corresponding to the material type and the environmental features; Inputting the input feature set into a pre-trained parameter configuration model, predicting optimal multispectral image acquisition parameters, and obtaining model prediction results; Adaptively adjusting the multispectral image acquisition parameters according to the prediction results of the model; The detection method further comprises a training step of a parameter configuration model, wherein the training step comprises: Acquire and preprocess a historical sample feature set; the historical sample feature set includes optimal multispectral image acquisition parameters for different materials under different environments and corresponding image quality indicators; Perform feature extraction on the preprocessed historical sample feature set to obtain a basic feature vector set; Dividing the basic feature vector set into training samples and verification samples; Inputting the training samples into a pre-built gradient boosting tree model for initial training, optimizing model parameters, and obtaining an initial training parameter configuration model; The parameter configuration model is verified and model parameters are corrected based on the verification sample until the loss function of the model meets a preset condition or the number of model iterations reaches a preset number, thereby obtaining the trained parameter configuration model. 2. The intelligent cloth inspection method according to claim 1 is characterized in that the step of performing pixel-by-pixel analysis on the pre-processed multispectral image data according to preset logical rules and preset feature thresholds comprises: Perform image segmentation based on the preprocessed multispectral image data to obtain a plurality of homogeneous regions that are initially divided; Extract the spectral features of each pixel in each spectral band in multiple homogeneous areas to obtain the single pixel features of each pixel; Calculate the difference in spectral characteristics between each pixel and its adjacent pixels to obtain the neighborhood characteristics of each pixel; According to the single pixel features and neighborhood features, a comprehensive feature vector of each pixel is obtained by weighted fusion; Based on the preset logical rules, determine whether the comprehensive feature vector of each pixel exceeds the preset feature threshold; if so, mark the pixel as a defective pixel; if not, mark the pixel as a normal pixel; Obtaining a defective pixel set according to the plurality of defective pixel points; Based on an image processing algorithm, defining a defect boundary of a defect area according to the defect pixel set; The multispectral image data is divided into a normal area and a defect area according to the defect boundary. 3. The intelligent cloth inspection method according to claim 2 is characterized in that the step of analyzing the time series images to obtain the moving trajectory and change trend of the defective area on the cloth to be inspected and obtaining the dynamic characteristics of the defective area comprises: Performing image registration on the time series images of the defective area; Extract defect feature vectors based on the registered time series images and construct a time series of defect feature vectors; Based on the dynamic time warping algorithm, the similarity of defect feature vectors between adjacent frames is compared according to the time series of defect feature vectors, and the same defects in consecutive frames are matched; According to the same defects in consecutive frames, a continuous moving trajectory of the defect feature vector is constructed; Extracting defect variation parameters according to the continuous moving trajectory and identifying the variation trend of the defect based on a time series analysis algorithm; According to the continuous movement trajectory and the change trend, the dynamic characteristics of the defect area are extracted; wherein the dynamic characteristics include diffusion rate, color change and movement path. 4. The intelligent cloth inspection method according to any one of claims 1 to 3, characterized in that after the step of identifying the normal area and the defective area on the cloth to be inspected, it also includes: Determine a normal area on the fabric to be tested as a sample test area, and obtain a sample coordinate set of the sample test area; Perform near-infrared light reflection according to the sample coordinate set to obtain near-infrared spectrum data of the sample test area; Preprocessing the near infrared spectrum data of the sample test area; Inputting the pre-processed near-infrared spectral data into a pre-built machine learning model to obtain a chemical residue analysis result of the sample test area; wherein the chemical residue analysis result includes the type and concentration value of the chemical residue at each sample point in the sample test area; Based on the preset degradable substance database, the degradability level corresponding to the fabric to be tested is evaluated according to the chemical residue analysis result and sent to the management terminal. 5. The intelligent cloth inspection method according to any one of claims 1 to 3, characterized in that after the step of identifying the normal area and the defective area on the cloth to be inspected, it also includes: Determine a normal area on the fabric to be tested as a sample test area, and obtain a sample coordinate set of the sample test area; Perform near-infrared light reflection according to the sample coordinate set to obtain near-infrared spectrum data of the sample test area; Preprocessing the near infrared spectrum data of the sample test area; The preprocessed spectral data are processed by first-order and second-order derivatives, and potential characteristic peaks are identified based on the peak detection algorithm; Filter the noise peaks in the potential characteristic peaks to obtain chemical characteristic peaks; Matching the chemical characteristic peaks with a preset chemical residue spectral characteristic library to determine the type of chemical residue at each sample point; Calculate the concentration value corresponding to the chemical residue type at each sample point based on the standard curve method; According to the chemical residue types and concentration values of each sample point, the chemical residue analysis results are obtained; Based on the preset degradable substance database, the degradability level corresponding to the fabric to be tested is evaluated according to the chemical residue analysis result and sent to the management terminal. 6. An intelligent cloth inspection system, characterized in that the inspection system comprises: A data acquisition module, used to acquire physical property data and spectral response data of the fabric to be tested; A material type identification module, used to determine the material type of the fabric to be detected according to the physical property data and the spectral response data; An acquisition parameter adjustment module, used for adaptively adjusting the multispectral image acquisition parameters according to the material type; A multispectral image data acquisition module, used for performing multispectral irradiation on the fabric to be detected and acquiring image data in different spectral bands based on the multispectral image acquisition parameters to obtain multispectral image data; An image preprocessing module, used for performing image preprocessing on the multispectral image data; An image analysis module, used to perform pixel-by-pixel analysis on the preprocessed multispectral image data according to preset logic rules and preset feature thresholds, and identify normal areas and defective areas of the multispectral image data; A marking module, used for marking the defective area; A time series image acquisition module, used for continuously photographing the defect area to obtain a time series image of the defect area; A dynamic characteristic generation module, used for analyzing the time series images to obtain the moving track and change trend of the defective area on the fabric to be detected, and obtaining the dynamic characteristics of the defective area; A defect detection module, used to determine the corresponding defect type according to the dynamic characteristics of the defect area, obtain the defect detection result and send it to the management terminal; The acquisition parameter adjustment module is configured as follows: Acquire current environmental data in real time and perform data preprocessing; Extracting environmental features corresponding to the preprocessed current environmental data; constructing an input feature set based on the category features corresponding to the material type and the environmental features; Inputting the input feature set into a pre-trained parameter configuration model, predicting optimal multispectral image acquisition parameters, and obtaining model prediction results; Adaptively adjusting the multispectral image acquisition parameters according to the prediction results of the model; The training step of the parameter configuration model includes: Acquire and preprocess a historical sample feature set; the historical sample feature set includes optimal multispectral image acquisition parameters for different materials under different environments and corresponding image quality indicators; Perform feature extraction on the preprocessed historical sample feature set to obtain a basic feature vector set; Dividing the basic feature vector set into training samples and verification samples; Inputting the training samples into a pre-built gradient boosting tree model for initial training, optimizing model parameters, and obtaining an initial training parameter configuration model; The parameter configuration model is verified and model parameters are corrected based on the verification sample until the loss function of the model meets a preset condition or the number of model iterations reaches a preset number, thereby obtaining the trained parameter configuration model. 7. A computer device, characterized in that it comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 5 when executing the program. 8. A computer-readable storage medium, characterized in that it stores a computer program that can be loaded by a processor and execute the method according to any one of claims 1 to 5.