A color change characteristic analysis method and system for highly dyed polyester fabrics

Abstract

The present invention relates to the technical field of textile fabric image processing, and specifically to a method and system for analyzing the color change characteristics of highly dyed polyester fabric. First, the present invention tests highly dyed polyester fabric samples under different test conditions and collects multispectral images of the fabric samples before and after the test; secondly, the multispectral images before and after the test are preprocessed using an image processing algorithm to restore the true color; then, the color difference of the fabric samples before and after the test is calculated, and the color change level is set; then, the color change characteristics of the fabric sample image are extracted, and the potential characteristics of the fabric sample image are extracted through a GAN model; finally, the SVM model is used to classify the color change level and predict the color difference value of the fabric sample image after the test, and the influence of different factors on the color change of the highly dyed polyester fabric is analyzed. The present invention combines the GAN‑SVM model to analyze the color change characteristics of the highly dyed polyester fabric, and achieves accurate color change level classification and color difference value prediction.

Description

Technical Field

[0001] The invention relates to the technical field of textile fabric image processing, and in particular to a method and system for analyzing color change characteristics of highly dyed polyester fabric. Background Art

[0002] High-dyed polyester fabric is a special polyester textile material, usually a polyester fabric treated with high temperature and high pressure dyeing process, which has good color fastness and dyeing effect. In the modern textile industry, high-dyed polyester fabric is favored for its bright colors and good durability, and is widely used in many fields, such as high-end fashion, sportswear, curtains and bedding. However, due to its complex dyeing process and high pigment content, high-dyed polyester fabric is prone to color change during production, transportation and use, affecting its appearance and quality. Therefore, how to accurately evaluate the color characteristics of high-dyed polyester fabric has become an important topic in the textile industry.

[0003] In the analysis of color change characteristics of highly dyed polyester fabrics, the application of artificial intelligence technology has made some progress, such as using support vector machines to distinguish and identify different color components, and evaluating dyeing effects and color change by measuring indicators such as K / S value, color fastness to rubbing and color fastness to light. However, existing color change characteristic analysis methods for highly dyed polyester fabrics usually require a large amount of training data and computing resources, especially when dealing with complex dyeing process factors. Otherwise, they may not be able to fully reflect the color change characteristics of highly dyed polyester fabrics, thereby affecting the accuracy and reliability of the color change analysis results.

[0004] Therefore, a color change characteristic analysis method and system of highly dyed polyester fabrics are proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for analyzing the color change characteristics of highly dyed polyester fabrics. The present invention first tests highly dyed polyester fabric samples under different test conditions and collects multispectral images of the fabric samples before and after the test; secondly, the multispectral images before and after the test are preprocessed by using an image processing algorithm to restore the color difference; then, the color difference of the fabric samples before and after the test is calculated and the color change level is set; then, the color change characteristics of the fabric sample images after the test are extracted, and the potential characteristics of the fabric sample images after the test are extracted by a GAN model; finally, the SVM model is used to classify the color change levels of the fabric sample images after the test, and the influence of different factors on the color change of the highly dyed polyester fabric is analyzed.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for analyzing color change characteristics of highly dyed polyester fabrics, comprising:

[0008] Multispectral images of highly dyed polyester fabric samples were collected, and the uniformity of the images was ensured by setting appropriate lighting conditions to obtain the original fabric sample images.

[0009] Preferably, different test conditions are set to process the highly dyed polyester fabric sample, simulating the influence of external factors that the highly dyed polyester fabric may suffer during dyeing and actual use, and obtaining the tested highly dyed polyester fabric sample; the test conditions include: dyeing temperature conditions, fixation time conditions, auxiliary agent conditions, exposure conditions, friction conditions, sweat stain conditions and washing conditions;

[0010] Wherein, the dyeing temperature conditions include: setting different temperatures for dyeing to achieve different dyeing effects;

[0011] The fixing time condition includes: setting different fixing times for testing to achieve different fixing effects; the auxiliary agent condition includes: adding different types of auxiliary agents for dyeing;

[0012] The exposure conditions include: exposing the highly dyed polyester fabric sample to natural light for testing, and setting different exposure times;

[0013] The friction conditions include: using a standard friction meter to perform a friction test on the highly dyed polyester fabric sample, setting different friction times;

[0014] The sweat stain condition includes: testing the highly dyed polyester fabric sample using artificial sweat;

[0015] The water washing conditions include: performing water washing tests using different washing methods and washing times.

[0016] Preferably, a multispectral image of the test highly dyed polyester fabric sample is collected to ensure that the lighting conditions are consistent with those before the test, so as to obtain an image of the test fabric sample.

[0017] Preferably, the original fabric sample image and the test fabric sample image are preprocessed using an image processing algorithm to obtain a standard original fabric sample image and a standard test fabric sample image; the preprocessing includes: image denoising, image normalization, geometric correction and color correction;

[0018] Wherein, the image denoising comprises: using Gaussian filtering to remove noise in the test fabric sample image;

[0019] The image normalization includes: converting the RGB color space of the original fabric sample image and the test fabric sample image into the Lab color space so that the color change is not affected by the change of illumination and is more suitable for color difference calculation; adjusting the brightness distribution range in the image to the standard range through the brightness channel L of the Lab image;

[0020] The geometric correction includes: rotating, translating and scaling the original fabric sample image and the test fabric sample image, finding the positions of feature points between multiple images to determine the relative positions of the images, and determining and cutting out the fabric area containing the feature points through an image segmentation algorithm to remove irrelevant background information;

[0021] The color correction includes: converting the RGB color space of the original fabric sample image and the test fabric area image into the Lab color space, using a color equalization algorithm to adjust the brightness of different color channels so as to restore the color of the original fabric sample image, and converting the adjusted Lab component back to the RGB color space to obtain the standard original fabric sample image and the standard test fabric sample image.

[0022] Preferably, the color difference between the standard original fabric sample image and the standard test fabric sample image is calculated, and the color change level is set; the color difference calculation includes: calculating the color difference between the standard original fabric sample image and the standard test fabric sample image by using the color difference formula CIEDE2000; the specific color difference calculation formula is:

[0023] ;

[0024] in, is color difference; For brightness difference; is the chromaticity difference; For hue difference; is the brightness adjustment coefficient; is the chromaticity adjustment coefficient; is the hue adjustment coefficient; is the brightness scaling factor; is the chroma scaling factor; is the hue scaling factor; is the interaction term;

[0025] The color change level is set according to the color difference The value of is used to set an automated color change grading system for quickly classifying different degrees of color change; the color change grades include: There is no perceptible color difference; Slight color difference; Moderate color difference; Severe color difference.

[0026] Preferably, color change features are extracted based on the standard test fabric sample image, and the color change features include: a first color feature, a first texture feature, and a multi-spectral feature;

[0027] The extraction of the first color feature includes: calculating the average values ​​of the brightness L, the red-green color a and the yellow-blue color b in the entire image area in the Lab color space to obtain the Lab mean feature; Analyze and extract the value The mean and variance of the values ​​were used to evaluate the degree of fabric color change under different test conditions. Features; calculate the contrast through the brightness channel L to describe the sensitivity difference of the fabric surface and obtain the contrast feature; combine the Lab mean feature, feature and contrast feature to obtain the first color feature;

[0028] The extraction of the first texture feature includes: extracting the energy, contrast and entropy of the fabric texture in the standard original fabric sample image and the standard test fabric sample image through a gray level co-occurrence matrix to obtain gray level co-occurrence matrix features; using local binary pattern LBP statistics to calculate the LBP value distribution of all pixels in the fabric image to form a feature histogram; calculating the mean and variance of the feature histogram to obtain LBP features; combining the gray level co-occurrence matrix features and the LBP features as the first texture feature;

[0029] The extraction of the multispectral features includes: extracting pixel values ​​of different bands, including red, green, blue and near-infrared bands, from the standard original fabric sample image and the standard test fabric sample image; calculating the average reflectivity, maximum reflectivity, minimum reflectivity and reflectivity ratio of each band to obtain the multispectral features.

[0030] Preferably, a simulated test image is generated according to different test conditions by a GAN model, and the GAN potential features of the generated simulated test image are extracted; the implementation method of the GAN model includes: inputting the standard original fabric sample image and the standard test fabric sample image as training data of the GAN model; generating a simulated test image similar to the test fabric sample image according to different test conditions by a generator, thereby capturing potential features; judging whether the simulated test image conforms to the input standard test fabric sample image by a discriminator;

[0031] The specific training process is: fix the discriminator, train the generator, and generate a simulated test image; fix the generator, train the discriminator, and judge the authenticity of the simulated test image generated by the generator; repeat the above two steps, and alternately train the generator and the discriminator through the GAN loss function to generate a simulated test image that is most similar to the standard test fabric sample image; wherein the GAN loss function includes: generator loss and discriminator loss; the generator loss is to maximize the probability that the discriminator considers the generated simulated test image to be real, and the specific formula is:

[0032] ;

[0033] in, is the loss function of the generator; is the generator network; is random noise; To generate binary probabilities for simulated test images;

[0034] The discriminator loss is to maximize the accuracy of classifying the standard test fabric sample image and the generated simulated test image. The specific formula is:

[0035] ;

[0036] in, is the loss function of the discriminator;

[0037] After the GAN model training is completed, the potential features of the standard test fabric sample image are extracted through the generator and the discriminator. The specific process is: the standard test fabric sample image is input into the trained middle layer of the generator, the feature vector of the middle layer of the generator network is extracted, and the potential features of the generator are obtained; the potential features of the generator include: the texture, color and structure changes of the standard test fabric sample image;

[0038] Input the standard test fabric sample image and the simulated test image into the trained discriminator, extract the feature vector of the high-level convolutional layer of the discriminator, and obtain the discriminator potential features; the discriminator potential features include: overall color distribution and large-scale texture changes;

[0039] Combining the generator latent features and the discriminator latent features, the GAN latent features are obtained, including: second texture features, second color features, structural changes, overall color distribution and large-scale texture changes; combining the GAN latent features with the color change features to obtain a multi-dimensional feature vector.

[0040] Preferably, the color change grade of the standard test fabric sample images is classified using the SVM model according to the multidimensional feature vector to analyze the influence of different factors on the color change of highly dyed polyester fabrics; the specific classification and prediction process is: inputting the multidimensional feature vector into the SVM model, and annotating the standard test fabric sample images according to the color change grade; dividing the standard test fabric sample images into a training set, a validation set and a test set; selecting the RBF kernel as the kernel function, and training the SVM model and optimizing the parameters of the SVM model through cross-validation on the validation set; using the trained SVM model to classify the color change grade and predict the color difference value of the standard test fabric sample images in the test set to analyze the influence of different factors on the color change of highly dyed polyester fabrics; evaluating the color change grade classification result of the SVM model by calculating the accuracy; and evaluating the color difference value prediction result of the SVM model by calculating the mean square error.

[0041] A color change characteristic analysis system for highly dyed polyester fabrics, comprising:

[0042] A multispectral image acquisition module is used to acquire a multispectral image of a highly dyed polyester fabric sample to obtain an original fabric sample image; acquire a multispectral image of the test highly dyed polyester fabric sample to capture the color change under different test conditions to obtain a test fabric sample image;

[0043] Preferably, a multiple condition test module is used to set different test conditions to process the highly dyed polyester fabric sample, simulate the external influence factors that the highly dyed polyester fabric may suffer during dyeing and actual use, and obtain the tested highly dyed polyester fabric sample; the test conditions include: dyeing temperature conditions, fixation time conditions, auxiliary agent conditions, exposure conditions, friction conditions, sweat stain conditions and washing conditions;

[0044] Wherein, the dyeing temperature conditions include: setting different temperatures for dyeing to achieve different dyeing effects;

[0045] The color fixing time condition includes: setting different color fixing times for testing to achieve different color fixing effects;

[0046] The auxiliary agent conditions include: adding different types of auxiliary agents for dyeing;

[0047] The exposure conditions include: exposing the highly dyed polyester fabric sample to natural light for testing, and setting different exposure times;

[0048] The friction conditions include: using a standard friction meter to perform a friction test on the highly dyed polyester fabric sample, setting different friction times;

[0049] The sweat stain condition includes: testing the highly dyed polyester fabric sample using artificial sweat;

[0050] The water washing conditions include: performing water washing tests using different washing methods and washing times.

[0051] Preferably, a preprocessing module is used to preprocess the original fabric sample image and the test fabric sample image to restore the visual color of the highly dyed polyester fabric sample, and obtain a standard original fabric sample image and a standard test fabric sample image; the preprocessing includes: image denoising, image normalization, geometric correction and color correction;

[0052] Wherein, the image denoising comprises: using Gaussian filtering to remove noise in the original fabric sample image and the test fabric sample image;

[0053] The image normalization includes: converting the RGB color space of the original fabric sample image and the test fabric sample image into the Lab color space so that the color change is not affected by the change of illumination and is more suitable for color difference calculation; adjusting the brightness distribution range in the image to the standard range through the brightness channel L of the Lab image;

[0054] The geometric correction includes: rotating, translating and scaling the original fabric sample image and the test fabric sample image, finding the positions of feature points between multiple images to determine the relative positions of the images, and determining and cutting out the fabric area containing the feature points through an image segmentation algorithm to remove irrelevant background information;

[0055] The color correction includes: converting the RGB color space of the original fabric sample image and the test fabric area image into the Lab color space, using a color equalization algorithm to adjust the brightness of different color channels so as to restore the color of the original fabric sample image, and converting the adjusted Lab component back to the RGB color space to obtain the standard original fabric sample image and the standard test fabric sample image.

[0056] Preferably, the color difference calculation and color change level setting module is used to calculate the color difference between the standard original fabric sample image and the standard test fabric sample image, and set the color change level; the color difference calculation includes: calculating the color difference between the standard original fabric sample image and the standard test fabric sample image by using the color difference formula CIEDE2000; the specific color difference calculation formula is:

[0057] ;

[0058] in, is color difference; For brightness difference; is the chromaticity difference; For hue difference; is the brightness adjustment coefficient; is the chromaticity adjustment coefficient; is the hue adjustment coefficient; is the brightness scaling factor; is the chroma scaling factor; is the hue scaling factor; is the interaction term;

[0059] The color change level is set according to the color difference The value of is used to set an automated color change grading system for quickly classifying different degrees of color change; the color change grades include: There is no perceptible color difference; Slight color difference; Moderate color difference; Severe color difference.

[0060] Preferably, the color change feature extraction module is used to extract color change features according to the standard test fabric sample image, and the color change features include: a first color feature, a first texture feature and a multi-spectral feature;

[0061] The extraction of the first color feature includes: calculating the average values ​​of the brightness L, the red-green color a and the yellow-blue color b in the entire image area in the Lab color space to obtain the Lab mean feature; Analyze and extract the value The mean and variance of the values ​​were used to evaluate the degree of fabric color change under different test conditions. Characteristics; the contrast is calculated through the brightness channel L to describe the sensitivity difference of the fabric surface and obtain the contrast characteristics; the calculation formula of the contrast is:

[0062] ;

[0063] in, is the contrast; is the brightness channel; is the maximum value function; is the minimum function;

[0064] Combined with the Lab mean characteristics, feature and contrast feature to obtain the first color feature;

[0065] The extraction of the first texture feature includes: extracting the energy, contrast and entropy of the fabric texture in the standard original fabric sample image and the standard test fabric sample image through a gray level co-occurrence matrix to obtain gray level co-occurrence matrix features; using local binary pattern LBP statistics to calculate the LBP value distribution of all pixels in the fabric image to form a feature histogram; calculating the mean and variance of the feature histogram to obtain LBP features; combining the gray level co-occurrence matrix features and the LBP features as the first texture feature;

[0066] The extraction of the multispectral features includes: extracting pixel values ​​of different bands, including red, green, blue and near-infrared bands, from the standard original fabric sample image and the standard test fabric sample image; calculating the average reflectivity, maximum reflectivity, minimum reflectivity and reflectivity ratio of each band to obtain the multispectral features.

[0067] Preferably, the potential feature extraction module is used to generate simulated test images according to different test conditions through the GAN model, extract the GAN potential features of the generated simulated test images; combine the GAN potential features with the color change features to obtain a multi-dimensional feature vector; the implementation method of the GAN model includes: inputting the standard original fabric sample image and the standard test fabric sample image as training data of the GAN model; generating simulated test images similar to the test fabric sample image according to different test conditions through the generator, thereby capturing potential features; judging whether the simulated test image conforms to the input standard test fabric sample image through the discriminator;

[0068] The specific training process is: fix the discriminator, train the generator, and generate a simulated test image; fix the generator, train the discriminator, and judge the authenticity of the simulated test image generated by the generator; repeat the above two steps, and alternately train the generator and the discriminator through the GAN loss function to generate a simulated test image that is most similar to the standard test fabric sample image; wherein the GAN loss function includes: generator loss and discriminator loss; the generator loss is to maximize the probability that the discriminator considers the generated simulated test image to be real, and the specific formula is:

[0069] ;

[0070] in, is the loss function of the generator; is the generator network; is random noise; To generate binary probabilities for simulated test images;

[0071] The discriminator loss is to maximize the accuracy of classifying the standard test fabric sample image and the generated simulated test image. The specific formula is:

[0072] ;

[0073] in, is the loss function of the discriminator;

[0074] After the GAN model training is completed, the potential features of the standard test fabric sample image are extracted through the generator and the discriminator. The specific process is: the standard test fabric sample image is input into the trained middle layer of the generator, the feature vector of the middle layer of the generator network is extracted, and the potential features of the generator are obtained; the potential features of the generator include: the texture, color and structure changes of the standard test fabric sample image;

[0075] Input the standard test fabric sample image and the simulated test image into the trained discriminator, extract the feature vector of the high-level convolutional layer of the discriminator, and obtain the discriminator potential features; the discriminator potential features include: overall color distribution and large-scale texture changes;

[0076] Combining the generator latent features and the discriminator latent features, the GAN latent features are obtained, including: second texture features, second color features, structural changes, overall color distribution and large-scale texture changes; combining the GAN latent features with the color change features to obtain a multi-dimensional feature vector.

[0077] Preferably, the color change classification and color difference prediction module uses an SVM model to classify the color change level of the standard test fabric sample image according to the multidimensional feature vector, so as to analyze the influence of different factors on the color change of highly dyed polyester fabric; the specific process of SVM model construction and classification prediction is: inputting the multidimensional feature vector into the SVM model, and annotating the standard test fabric sample image according to the color change level; dividing the standard test fabric sample image into a training set, a validation set and a test set; selecting the RBF kernel as the kernel function to construct the SVM model, and training the SVM model and optimizing the parameters of the SVM model through cross-validation on the validation set; using the trained SVM model to classify the color change level and predict the color difference value of the standard test fabric sample image in the test set, so as to analyze the influence of different factors on the color change of highly dyed polyester fabric; evaluating the color change level classification result of the SVM model by calculating the accuracy; and evaluating the color difference value prediction result of the SVM model by calculating the mean square error.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] 1. The present invention adopts different test conditions to test highly dyed polyester fabrics, which can fully simulate various external influences that the fabrics may be subjected to during dyeing, processing and actual use, including different dyeing temperatures, fixation time, use of auxiliaries, exposure, friction, sweat stains and washing conditions; this systematic test helps to deeply understand the color change and stability of the fabrics under different conditions, so as to evaluate the actual application performance of the fabrics; at the same time, this multi-condition test can also help manufacturers optimize the dyeing process, select the best auxiliary formulation and processing parameters, and improve the color fastness and service life of the product; it provides high-quality training data for the subsequent multi-dimensional feature extraction of highly dyed polyester fabrics and thus analyzes the color change of highly dyed polyester fabrics.

[0080] 2. The present invention extracts the color change characteristics of highly dyed polyester fabrics and extracts its potential characteristics through the GAN model, which can deeply analyze the color, texture and structural changes of fabrics under different conditions; the color change characteristics provide intuitive quantitative information, such as the numerical description of color changes and the pattern of texture changes, which provides a reliable basis for evaluating fabric quality and color fastness; and the GAN model further explores the potential characteristics of fabrics, including complex color distribution laws and multi-scale texture change characteristics, which improves the recognition ability of the model of the present invention for color change characteristics; this combination provides a solid foundation for fabric production quality monitoring, process optimization and new material development, and promotes the automation and accuracy of the subsequent SVM model for the performance analysis of highly dyed polyester fabrics.

[0081] 3. The present invention combines the GAN model and the SVM model to classify the color change grade and predict the color difference value of highly dyed polyester fabrics, wherein the GAN model extracts the potential color, texture and structural characteristics of the fabric under complex conditions through adversarial learning of the generator and the discriminator, and captures the tiny and deep color change patterns; the SVM model uses these multidimensional features for classification and regression, accurately divides the color change grade and predicts the color difference value; the combination of the two not only makes up for the problem of insufficient data in the actual analysis process, but also enhances the adaptability of the model of the present invention to complex color change features, thereby improving the accuracy of the color change grade classification of highly dyed polyester fabrics and the accuracy of the color difference value prediction; in actual industrial scenarios, by combining the GAN model and the SVM model, accurate color change analysis of highly dyed polyester fabrics can be achieved, thereby monitoring the quality of highly dyed polyester fabrics. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 A schematic flow chart of a method for analyzing color change characteristics of a highly dyed polyester fabric provided by an embodiment of the present invention;

[0083] Figure 2 A schematic structural diagram of a color change characteristic analysis system for a highly dyed polyester fabric provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0084] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0085] In the modern textile industry, highly dyed polyester fabrics are favored for their bright colors and good durability and are widely used in many fields, such as high-end fashion, sportswear, curtains and bedding. However, due to its complex dyeing process and high pigment content, highly dyed polyester fabrics are prone to color change during production, transportation and use, affecting their appearance and quality. By accurately identifying and quantifying the color difference and texture changes of fabrics during dyeing and use, it can help optimize the dyeing process, improve fabric formulations and improve product stability. At the same time, analyzing the color change characteristics provides a scientific basis for predicting the durability of fabrics in different environments, and provides technical support for product development and quality management of fabric manufacturers, thereby enhancing market competitiveness and meeting consumer demand for high-quality textiles.

[0086] The present invention proposes a color change characteristic analysis method for highly dyed polyester fabrics, which realizes the classification of color change grades and color difference value prediction of highly dyed polyester fabrics. This method is applied to a color change characteristic analysis system for highly dyed polyester fabrics. The flowchart of the specific method and system is shown in Figure 1 and Figure 2 In order to illustrate that the method and system of the present invention can play a role in classifying the color change levels of highly dyed polyester fabrics and predicting the color difference values, the effectiveness of the present invention will be described from two examples below.

[0087] Embodiment 1

[0088] In the specific implementation process of this embodiment, the method of the present invention is implemented by a specific system module. Figure 1The specific steps of the method proposed in the present invention are described in detail; the specific steps include: S10. Testing highly dyed polyester fabric samples under different test conditions, and collecting multispectral images of the fabric samples before and after the test; S20. Preprocessing the multispectral images before and after the test using an image processing algorithm; S30. Calculating the color difference of the fabric samples before and after the test, and setting the color change level; S40. Extracting the color change features of the fabric sample images after the test; S50. Extracting the potential features of the fabric sample images after the test through a GAN model, and combining them with the color change features to obtain a multidimensional feature vector; S60. Based on the multidimensional feature vector, using an SVM model to classify the color change level and predict the color difference value of the fabric sample images after the test, and analyzing the influence of different factors on the color change of highly dyed polyester fabrics.

[0089] The multispectral image of the highly dyed polyester fabric of the present invention is composed of Figure 2 The multispectral image acquisition module in the system shown is acquired through a multispectral camera, corresponding to the above step S10, to obtain the original fabric sample image; the acquisition process of the original fabric sample image is as follows: the highly dyed polyester fabric sample is placed flat on a standard background to avoid wrinkles or irregular sample shapes affecting image acquisition; in order to ensure that the light source in the acquisition environment is uniform and stable, white LED lights are set at 45° on the upper left and upper right of the highly dyed polyester fabric sample to evenly illuminate the fabric sample to avoid interference from natural light or other unstable light sources; multispectral camera calibration is performed regularly to ensure that the camera responds consistently to different wavelengths; the spectral range is set, including the ultraviolet band (300-400nm), the visible light band (400-700nm) and the near infrared band (700-1000nm); according to the predetermined band, the image of each band is acquired in turn, and each band image should be clear and noise-free, and can accurately reflect the spectral reflectance of the band;

[0090] After collecting images for each band, check the image quality in time to ensure that the image is clear and noise-free; if image quality problems are found, adjust the camera settings or lighting conditions in time and re-collect.

[0091] Preferably, by Figure 2 The multiple condition test modules in the system set different test conditions to process the highly dyed polyester fabric samples, simulate the influence of external factors that the highly dyed polyester fabrics may suffer during dyeing and actual use, and obtain the tested highly dyed polyester fabric samples; the specific test process is: randomly select fabric samples with different color change levels from the highly dyed polyester fabric sample library, and process the fabrics according to the predetermined test conditions; the test conditions include: dyeing temperature conditions, color fixation time conditions, auxiliary agent conditions, exposure conditions, friction conditions, sweat stain conditions and washing conditions;

[0092] The dyeing temperature conditions include: setting different temperatures (80°C, 100°C and 130°C) for dyeing to simulate the dyeing process of the fabric at different temperatures;

[0093] The fixing time condition includes: setting different fixing times (30 minutes, 60 minutes and 90 minutes) for testing to achieve different fixing effects;

[0094] The auxiliary agent conditions include: adding different types of auxiliary agents (dyeing accelerator, leveling agent and dispersant) for dyeing to simulate the dyeing effect of fabrics under different auxiliary agents;

[0095] The exposure conditions include: setting different exposure times (1 hour, 3 hours and 6 hours), exposing the highly dyed polyester fabric sample to natural light for testing, simulating the sunlight radiation to which the highly dyed polyester fabric is exposed during wearing;

[0096] The friction conditions include: setting different friction times (50 times, 100 times and 200 times), using a standard friction meter to perform friction tests on the highly dyed polyester fabric sample, simulating the friction damage that the fabric may encounter during use;

[0097] The sweat stain condition includes: using artificial sweat to test the highly dyed polyester fabric sample to simulate the color change of the fabric after human sweating; the washing condition includes:

[0098] Washing tests were conducted using different washing methods and washing times (hand washing 1, 3, 5 times and machine washing 1, 3, 5 times) to simulate the effect of washing on fabric color;

[0099] After completing the test processing of the sample, a multispectral camera is used to collect multispectral images of the tested high-dyed polyester fabric sample to obtain a test fabric sample image; wherein, the sample placement method, light source setting, and band selection involved in the collection process remain consistent with those before the test.

[0100] The embodiment of the present application performs multispectral image acquisition on highly dyed polyester fabric samples, records in detail the reflectance characteristics of highly dyed polyester fabrics in different bands, and captures the color and texture changes of the fabrics; the highly dyed polyester fabric samples are processed by setting different dyeing temperatures, fixation times, additives, exposure, friction, sweat stains and washing conditions to simulate the color change of the fabrics caused by different external environmental factors during dyeing and actual use; this step can comprehensively analyze the color change characteristics of highly dyed polyester fabrics through high-quality image acquisition, precise test condition setting and standardized sample processing, and provide reliable data support for subsequent color change feature analysis, feature extraction and color change grade classification.

[0101] Preferably, by Figure 2 The preprocessing module in the system preprocesses the original fabric sample image and the test fabric sample image to restore the visual color of the highly dyed polyester fabric sample, and corresponding to the above step S20, obtains a standard original fabric sample image and a standard test fabric sample image; the preprocessing includes: image denoising, image normalization, geometric correction and color correction; wherein the image denoising includes: using Gaussian filtering to remove noise from the test fabric sample image;

[0102] The image normalization includes: converting the RGB color space of the original fabric sample image and the test fabric sample image into the Lab color space so that the color change is not affected by the change of illumination and is more suitable for color difference calculation; adjusting the brightness distribution range in the image to the standard range through the brightness channel L of the Lab image;

[0103] The geometric correction includes: rotating, translating and scaling the original fabric sample image and the test fabric sample image, finding the positions of feature points between multiple images to determine the relative positions of the images, and determining and cutting out the fabric area containing the feature points through an image segmentation algorithm to remove irrelevant background information;

[0104] The color correction includes: converting the RGB color space of the original fabric sample image and the test fabric area image into the Lab color space, using a color equalization algorithm to adjust the brightness of different color channels so as to restore the color of the original fabric sample image, and converting the adjusted Lab component back to the RGB color space to obtain the standard original fabric sample image and the standard test fabric sample image.

[0105] The embodiment of the present application preprocesses the original fabric sample image and the test fabric sample image, with the aim of eliminating noise in the multispectral image acquisition process and the test process, standardizing the color and brightness of the multispectral image, aligning the multispectral images of the highly dyed polyester fabric taken from different viewing angles, cropping out the feature area, and eliminating the shooting color difference of the highly dyed polyester fabric image; through these preprocessing steps, a high-quality data foundation is provided for the subsequent color difference extraction and color change feature extraction, ensuring that the results of the subsequent color change grade classification and color difference prediction are more accurate.

[0106] Preferably, by Figure 2 The color difference calculation and color change level setting module in the system calculates the color difference between the standard original fabric sample image and the standard test fabric sample image, and sets the color change level, corresponding to the above step S30;

[0107] The color difference calculation includes: calculating the color difference between the standard original fabric sample image and the standard test fabric sample image by using the color difference formula CIEDE2000; the specific color difference calculation formula is:

[0108] ;

[0109] in, is color difference; For brightness difference; is the chromaticity difference; For hue difference; is the brightness adjustment coefficient; is the chromaticity adjustment coefficient; is the hue adjustment coefficient; is the brightness scaling factor; is the chroma scaling factor; is the hue scaling factor; is the interaction term;

[0110] The color change level is set according to the color difference The value of is used to set an automated color change grading system for quickly classifying different degrees of color change; the color change grades include: There is no perceptible color difference; Slight color difference; Moderate color difference; Severe color difference.

[0111] The embodiment of the present application calculates the color difference between the original fabric sample image and the test fabric sample image, quantifies the color difference, and sets a standardized color change level, so as to evaluate the color change of highly dyed polyester fabric under different test conditions; the implementation of this step provides a good foundation for the subsequent extraction of color change features, and at the same time provides a reliable basis for the color change quality evaluation of highly dyed polyester fabric and the improvement of dyeing performance during the dyeing process.

[0112] Preferably, by Figure 2 The color change feature extraction module in the system shown in the figure extracts color change features according to the standard test fabric sample image, corresponding to the above step S40; the color change features include: a first color feature, a first texture feature and a multi-spectral feature; wherein the extraction of the first color feature includes: in the Lab color space, calculating the average values ​​of the brightness L, the red-green color a and the yellow-blue color b in the entire image area, respectively expressed as , and , get the Lab mean feature; Analyze and extract the value The mean and variance of the values ​​were used to evaluate the degree of fabric color change under different test conditions. Features; the contrast is calculated through the brightness channel L to describe the sensitivity difference of the fabric surface and obtain the contrast feature; the calculation formula of the contrast in the color feature is:

[0113] ;

[0114] in, is the contrast in color features; is the brightness channel; is the maximum value function; is the minimum function;

[0115] Combined with the Lab mean characteristics, feature and contrast feature to obtain the first color feature;

[0116] The extraction of the first texture feature includes: extracting the energy, contrast, entropy and correlation of the fabric texture in the standard original fabric sample image and the standard test fabric sample image through a gray level co-occurrence matrix to obtain a gray level co-occurrence matrix feature; wherein the energy represents the uniformity of the fabric texture and reflects the degree of repeated patterns; the specific formula is:

[0117] ;

[0118] in, for energy; is the gray-level co-occurrence matrix; and is the gray value;

[0119] The contrast reflects the degree of contrast between the depth and lightness of the fabric texture, and the specific formula is:

[0120] ;

[0121] in, is the contrast in texture features;

[0122] The entropy represents the complexity of the fabric texture, and the calculation formula is:

[0123] ;

[0124] in, is entropy;

[0125] The LBP values ​​of all pixels in the fabric image are counted by using the local binary pattern LBP to form a feature histogram; the mean and variance of the feature histogram are calculated to obtain the LBP feature; the gray level co-occurrence matrix feature and the LBP feature are combined as the first texture feature;

[0126] The extraction of the multispectral features includes: extracting pixel values ​​of different bands from the standard original fabric sample image and the standard test fabric sample image, including: red, green, blue and near-infrared bands; calculating the average reflectivity of each band , maximum reflectivity , minimum reflectivity and reflectivity ratio , and , and obtain the multi-spectral features.

[0127] The embodiment of the present application extracts the first color feature, the first texture feature and the multi-spectral feature of the highly dyed polyester fabric, and combines them to obtain the color change feature; each type of feature reflects the specific situation of the color change from different dimensions, and the combination of multiple features comprehensively describes the color change behavior of the highly dyed polyester fabric under different test conditions, providing high-quality training data for the subsequent color change feature analysis, color change grade classification and color difference value prediction model.

[0128] Preferably, by Figure 2 The potential feature extraction module in the system shown corresponds to the above step S50, generates simulated test images according to different test conditions through the GAN model, and extracts the GAN potential features of the generated simulated test images;

[0129] The implementation method of the GAN model includes: inputting the standard original fabric sample image and the standard test fabric sample image as training data of the GAN model; generating a simulated test image similar to the test fabric sample image according to different test conditions by a generator, thereby capturing potential features; judging whether the simulated test image conforms to the input standard test fabric sample image by a discriminator;

[0130] The specific training process is: fix the discriminator, train the generator, and generate a simulated test image; fix the generator, train the discriminator, and judge the authenticity of the simulated test image generated by the generator; repeat the above two steps, and alternately train the generator and the discriminator through the GAN loss function to generate a simulated test image that is most similar to the standard test fabric sample image; wherein the GAN loss function includes: generator loss and discriminator loss; the generator loss is to maximize the probability that the discriminator considers the generated simulated test image to be real, and the specific formula is:

[0131] ;

[0132] in, is the loss function of the generator; is the generator network; is random noise; To generate binary probabilities for simulated test images;

[0133] The discriminator loss is to maximize the accuracy of classifying the standard test fabric sample image and the generated simulated test image. The specific formula is:

[0134] ;

[0135] in, is the loss function of the discriminator;

[0136] After the GAN model training is completed, the potential features of the standard test fabric sample image are extracted through the generator and the discriminator. The specific process is: the standard test fabric sample image is input into the trained middle layer of the generator, the feature vector of the middle layer of the generator network is extracted, and the generator potential features are obtained; the generator potential features include: the second texture features of the standard test fabric sample image , Second color feature and structural changes ;

[0137] The standard test fabric sample image and the simulated test image are input into the trained discriminator, and the feature vector of the high-level convolution layer of the discriminator is extracted to obtain the discriminator potential features; the discriminator potential features include: overall color distribution and large-scale texture variations ;

[0138] The generator latent feature and the discriminator latent feature are combined to obtain the GAN latent feature; the GAN latent feature is combined with the color change feature to obtain a multi-dimensional feature vector.

[0139] In order to facilitate the subsequent SVM model to classify the color change level and predict the color difference value of the highly dyed polyester fabric through the extracted multidimensional feature vectors, the extracted multidimensional feature vectors are listed in Table 1.

[0140] Table 1 Multidimensional feature vector

[0141]

[0142] In an embodiment of the present application, the generator and the discriminator of the GAN model compete with each other to learn the distribution characteristics of the original fabric sample image and the simulated fabric sample image; for the GAN potential feature extraction of highly dyed polyester fabric, the GAN model can capture the potential features through the middle layer of the generator, and evaluate the similarity between the generated simulated test image and the test fabric sample image through the high-level convolutional layer of the discriminator, thereby extracting the GAN potential features of the fabric image; these features integrate color, texture, structure, overall color change and large-scale texture change information, and are a high-level representation of the color change characteristics of highly dyed polyester fabrics; by combining the color change features and the GAN potential features, the color change characteristics of the highly dyed polyester fabric can be deeply analyzed to provide support for subsequent SVM model training and feature analysis.

[0143] Preferably, by Figure 2 The color change classification and color difference prediction module in the system shown corresponds to the above step S60, and classifies the color change level of the standard test fabric sample image using the SVM model according to the multidimensional feature vector to analyze the influence of different factors on the color change of the highly dyed polyester fabric; the construction of the SVM model includes: inputting the multidimensional feature vector into the SVM model, and annotating the color change level of the standard test fabric sample image according to the color change level, including 1 for no color difference, 2 for slight color difference, 3 for moderate color difference, and 4 for severe color difference; dividing the standard test fabric sample image into a 70% training set, a 15% validation set, and a 15% test set; selecting the RBF kernel function to construct the SVM model, and mapping the low-dimensional feature space to the high-dimensional feature space;

[0144] The training of the SVM model includes: setting hyperparameters C and , inputting the extracted multidimensional feature vector and the color change level into the SVM model, repeatedly iteratively adjusting and optimizing the parameters of the SVM model through grid search and cross-validation on the validation set, and selecting the optimal hyperparameter, wherein the optimal hyperparameter is the hyperparameter with the smallest error obtained by the SVM model on the validation set;

[0145] The trained SVM model is used to classify the color change level and predict the color difference value of the standard test fabric sample images in the test set to analyze the influence of different factors on the color change of highly dyed polyester fabrics. Table 2 shows the color change level classification and color difference value prediction results of 10 highly dyed polyester fabric samples by the SVM model through the multidimensional feature vector.

[0146] Table 2 SVM model for color change classification and color difference value prediction results of highly dyed polyester fabrics

[0147]

[0148] It can be seen from Table 2 that in the application of the SVM model in the classification of color change levels of highly dyed polyester fabrics, the color change levels of most samples are accurately classified; at the same time, the prediction results of the color difference values ​​of highly dyed polyester fabrics are close to the true values, indicating that the model has a good regression prediction effect on the color difference values ​​of highly dyed polyester fabrics; some samples (such as sample 8) are misclassified, which may be because the characteristics of such erroneous samples are close to the classification boundaries of different levels, making it difficult for the SVM model to distinguish.

[0149] The color change level classification results of the SVM model were evaluated by calculating the accuracy, precision and recall rate; the color difference value prediction results of the SVM model were evaluated by calculating the mean square error; Table 3 shows the performance evaluation of the SVM model.

[0150] Table 3 Performance evaluation of SVM model

[0151]

[0152] As shown in Table 3, the SVM model has an accuracy of 95.6%, a precision of 0.93, and a recall of 0.91 for the color change grade classification of highly dyed polyester fabrics, indicating that the SVM model performs well in the classification task of the color change grade of highly dyed polyester fabrics, and the classification results are relatively accurate; at the same time, the mean square error of the color difference prediction results of the SVM model for highly dyed polyester fabrics is 0.07, indicating that the SVM model has a small prediction error for the color difference value and an accurate prediction result.

[0153] In the embodiment of the present application, by constructing and training an SVM model to classify the color change grades and predict the color difference values ​​of highly dyed polyester fabrics, the complex color change characteristics in the fabric samples can be effectively captured, and the high accuracy of the color change grade classification is ensured, which is particularly suitable for polyester fabrics, which are materials with rich texture and color characteristics and complex changes; at the same time, the kernel function of the SVM model enables it to maintain a high prediction accuracy when facing the color difference and subtle changes of highly dyed polyester fabrics, thereby reducing the probability of classification errors; by using the SVM model to classify the color change grades and predict the color difference values ​​of highly dyed polyester fabrics, the quality inspection and color difference analysis of the fabrics can be performed accurately and quickly, thereby improving the production efficiency of highly dyed polyester fabrics and the quality control level of dyeing, reducing manual inspection errors, and achieving more accurate and automated production management of highly dyed polyester fabrics.

[0154] The embodiment of the present application realizes the color change grade classification and color difference prediction of highly dyed polyester fabrics with high accuracy through the comprehensive application of the test model, GAN model and SVM model proposed by the present invention. First, by setting different dyeing temperatures, fixation times, additives, exposure, friction, sweat stains and washing conditions to process the highly dyed polyester fabric samples, the color change caused by different external environmental factors during dyeing and actual use of the fabric is simulated, and the color change characteristics of the highly dyed polyester fabric can be comprehensively analyzed, providing reliable data support for subsequent color difference extraction and color change feature analysis. By pre-processing the highly dyed polyester fabric before and after the test to eliminate the noise in the multispectral image acquisition process and the test process, standardize the color and brightness of the multispectral image, align the multispectral images of the highly dyed polyester fabric taken from different perspectives, cut the required feature areas and eliminate the shooting color difference of the highly dyed polyester fabric image, a high-quality data basis is provided for the subsequent color difference extraction and color change feature extraction, ensuring that the results of the subsequent color change grade classification and color difference prediction are more accurate. By calculating the color difference of the fabric sample images before and after the test to quantify the color difference and setting the color change level, the color change of highly dyed polyester fabric under different test conditions can be evaluated, which provides a good basis for the subsequent extraction of color change features. The color change characteristics of highly dyed polyester fabrics are extracted to reflect the color change under different test conditions. The color change characteristics include the first color characteristics, the first texture characteristics and the multi-spectral characteristics. The combination of multiple characteristics comprehensively describes the color change behavior of highly dyed polyester fabrics under different test conditions, providing multi-dimensional and comprehensive training data for the subsequent color change feature analysis, color change level classification and color difference value prediction model. The GAN potential features of highly dyed polyester fabrics are obtained through adversarial training of the GAN model generator and the discriminator, including the second color feature, the second texture feature, the structural change, the overall color change and the large-scale texture change information, and are integrated with the color change features. The color change characteristics of highly dyed polyester fabrics can be deeply analyzed, providing support for the subsequent SVM model training and feature analysis. Finally, the present invention uses the SVM model to classify the color change grades and predict the color difference values ​​of highly dyed polyester fabrics according to the extracted multidimensional feature vectors, which can achieve accurate quality detection and color difference analysis of highly dyed polyester fabrics, improve the production efficiency of highly dyed polyester fabrics and the quality control level of dyeing, reduce manual inspection errors, and achieve more accurate and automated production management of highly dyed polyester fabrics.

[0155] Embodiment 2

[0156] The embodiment of the present invention provides a color change characteristic analysis system for highly dyed polyester fabrics, comprising:

[0157] A multispectral image acquisition module is used to acquire a multispectral image of a highly dyed polyester fabric sample to obtain an original fabric sample image; acquire a multispectral image of the test highly dyed polyester fabric sample to capture the color change under different test conditions to obtain a test fabric sample image;

[0158] A multiple condition test module is used to set different test conditions to process the highly dyed polyester fabric sample, simulate the external influence factors that the highly dyed polyester fabric may suffer during dyeing and actual use, and obtain the tested highly dyed polyester fabric sample; the test conditions include: dyeing temperature, fixation time, auxiliaries, exposure, friction, sweat stains and washing;

[0159] Among them, the dyeing temperature conditions include: setting different temperatures for dyeing to achieve different dyeing effects; the fixing time conditions include: setting different fixing times for testing to achieve different fixing effects; the auxiliary agent conditions include: adding different types of auxiliary agents for dyeing; the exposure conditions include: exposing the highly dyed polyester fabric sample to natural light for testing and setting different exposure times; the friction conditions include: using a standard friction meter to perform a friction test on the highly dyed polyester fabric sample and setting different friction times; the sweat stain conditions include: using artificial sweat to test the highly dyed polyester fabric sample; the washing conditions include: using different washing methods and washing times for washing tests.

[0160] A preprocessing module, used for preprocessing the original fabric sample image and the test fabric sample image to restore the visual color of the highly dyed polyester fabric sample, and obtain a standard original fabric sample image and a standard test fabric sample image; the preprocessing includes: image denoising, image normalization, geometric correction and color correction;

[0161] The color difference calculation and color change level setting module is used to calculate the color difference between the standard original fabric sample image and the standard test fabric sample image, and set the color change level; the color difference calculation includes: calculating the color difference between the standard original fabric sample image and the standard test fabric sample image by using the color difference formula CIEDE2000;

[0162] The color change level is set according to the color difference The value of is used to set an automated color change grading system for quickly classifying different degrees of color change; the color change grades include: There is no perceptible color difference; Slight color difference; Moderate color difference; Severe color difference.

[0163] A color change feature extraction module, used to extract color change features according to the standard test fabric sample image, wherein the color change features include: a first color feature, a first texture feature and a multi-spectral feature;

[0164] The extraction of the first color feature includes: calculating the average values ​​of the brightness L, the red-green color a and the yellow-blue color b in the entire image area in the Lab color space to obtain the Lab mean feature; Analyze and extract the value The mean and variance of the values ​​were used to evaluate the degree of fabric color change under different test conditions. Features; calculate the contrast through the brightness channel L to describe the sensitivity difference of the fabric surface and obtain the contrast feature; combine the Lab mean feature, feature and contrast feature to obtain the first color feature;

[0165] The extraction of the first texture feature includes: extracting the energy, contrast and entropy of the fabric texture in the standard original fabric sample image and the standard test fabric sample image through a gray level co-occurrence matrix to obtain gray level co-occurrence matrix features; using local binary pattern LBP statistics to calculate the LBP value distribution of all pixels in the fabric image to form a feature histogram; calculating the mean and variance of the feature histogram to obtain LBP features; combining the gray level co-occurrence matrix features and the LBP features as the first texture feature;

[0166] The extraction of the multispectral features includes: extracting pixel values ​​of different bands, including red, green, blue and near-infrared bands, from the standard original fabric sample image and the standard test fabric sample image; calculating the average reflectivity, maximum reflectivity, minimum reflectivity and reflectivity ratio of each band to obtain the multispectral features.

[0167] A potential feature extraction module is used to generate a simulated test image according to different test conditions through a GAN model, extract the GAN potential features of the generated simulated test image; combine the GAN potential features with the color change features to obtain a multi-dimensional feature vector; the implementation method of the GAN model includes: inputting the standard original fabric sample image and the standard test fabric sample image as training data of the GAN model; generating a simulated test image similar to the test fabric sample image according to different test conditions through a generator, thereby capturing potential features; judging whether the simulated test image conforms to the input standard test fabric sample image through a discriminator;

[0168] The specific training process is: fix the discriminator, train the generator, and generate a simulated test image; fix the generator, train the discriminator, and judge the authenticity of the simulated test image generated by the generator; repeat the above two steps, and alternately train the generator and the discriminator through the GAN loss function to generate a simulated test image that is most similar to the standard test fabric sample image; wherein the GAN loss function includes: generator loss and discriminator loss; the generator loss is to maximize the probability that the discriminator considers the generated simulated test image to be real, and the specific formula is:

[0169] ;

[0170] in, is the loss function of the generator; is the generator network; is random noise; To generate binary probabilities for simulated test images;

[0171] The discriminator loss is to maximize the accuracy of classifying the standard test fabric sample image and the generated simulated test image. The specific formula is:

[0172] ;

[0173] in, is the loss function of the discriminator;

[0174] After the GAN model training is completed, the potential features of the standard test fabric sample image are extracted through the generator and the discriminator. The specific process is: the standard test fabric sample image is input into the trained middle layer of the generator, the feature vector of the middle layer of the generator network is extracted, and the generator potential features are obtained; the generator potential features include: the texture, color and structural changes of the standard test fabric sample image; the standard test fabric sample image and the simulated test image are input into the trained discriminator, the feature vector of the high-level convolution layer of the discriminator is extracted, and the discriminator potential features are obtained; the discriminator potential features include: overall color distribution and large-scale texture changes;

[0175] Combining the generator latent features and the discriminator latent features, the GAN latent features are obtained, including: second texture features, second color features, structural changes, overall color distribution and large-scale texture changes; combining the GAN latent features with the color change features to obtain a multi-dimensional feature vector.

[0176] Preferably, the color change classification and color difference prediction module uses an SVM model to classify the color change level of the standard test fabric sample image according to the multidimensional feature vector, so as to analyze the influence of different factors on the color change of highly dyed polyester fabric; the specific process of SVM model construction and classification prediction is: inputting the multidimensional feature vector into the SVM model, and annotating the standard test fabric sample image according to the color change level; dividing the standard test fabric sample image into a training set, a validation set and a test set; selecting the RBF kernel as the kernel function to construct the SVM model, and training the SVM model and optimizing the parameters of the SVM model through cross-validation on the validation set; using the trained SVM model to classify the color change level and predict the color difference value of the standard test fabric sample image in the test set, so as to analyze the influence of different factors on the color change of highly dyed polyester fabric;

[0177] Table 4 shows the color change grade classification and color difference value prediction results of the SVM model for 10 highly dyed polyester fabric samples through the multidimensional feature vector.

[0178] Table 4 SVM model for color change classification and color difference value prediction results of highly dyed polyester fabrics

[0179]

[0180] It can be seen from Table 4 that the SVM model accurately classifies the color change levels of most samples in the application of color change level classification of highly dyed polyester fabrics; at the same time, the prediction results of the color difference values ​​of highly dyed polyester fabrics are close to the true value, indicating that the model has a good regression prediction effect on the color difference values ​​of highly dyed polyester fabrics.

[0181] The color change level classification result of the SVM model was evaluated by calculating the accuracy; the color difference value prediction result of the SVM model was evaluated by calculating the mean square error; Table 5 shows the performance evaluation of the SVM model.

[0182] Table 5 Performance evaluation of SVM model

[0183]

[0184] As shown in Table 5, the SVM model has an accuracy of 96.3%, a precision of 0.94, and a recall of 0.91 for the color change grade classification of highly dyed polyester fabrics, indicating that the SVM model performs well in the classification task of the color change grade of highly dyed polyester fabrics, and the classification results are relatively accurate; at the same time, the mean square error of the color difference prediction results of the SVM model for highly dyed polyester fabrics is 0.06, indicating that the SVM model has a small prediction error for the color difference value and an accurate prediction result.

[0185] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing the color change characteristics of highly dyed polyester fabrics, characterized in that: include: Collect multispectral images of highly dyed polyester fabric samples to obtain original fabric sample images; Different test conditions are set to process the highly dyed polyester fabric sample, simulating the influence of external factors on the highly dyed polyester fabric during dyeing and actual use, and obtaining the tested highly dyed polyester fabric sample; the test conditions include: dyeing temperature conditions, color fixation time conditions, auxiliary agent conditions, exposure conditions, friction conditions, sweat stain conditions and washing conditions; the dyeing temperature conditions include: setting different temperatures for dyeing to achieve different dyeing effects; the color fixation time conditions include: setting different color fixation times for testing to achieve different color fixation effects; the auxiliary agent conditions include: adding different types of auxiliary agents for dyeing; the exposure conditions include: exposing the highly dyed polyester fabric sample to natural light for testing, and setting different exposure times; the friction conditions include: using a standard friction meter to perform a friction test on the highly dyed polyester fabric sample, and setting different friction times; the sweat stain conditions include: using artificial sweat to test the highly dyed polyester fabric sample; the washing conditions include: using different washing methods to perform a washing test; Collecting a multispectral image of the test high-dyed polyester fabric sample to obtain a test fabric sample image; Preprocessing the original fabric sample image and the test fabric sample image using an image processing algorithm to obtain a standard original fabric sample image and a standard test fabric sample image; Calculating the color difference between the standard original fabric sample image and the standard test fabric sample image, and setting the color change level; Extracting color change features according to the standard test fabric sample image, wherein the color change features include: a first color feature, a first texture feature, and a multi-spectral feature; Generate simulated test images according to different test conditions through the GAN model, and extract the GAN potential features of the generated simulated test images, including: second texture features, second color features, structural changes, overall color distribution and large-scale texture changes; the implementation method of the GAN model includes: inputting the standard original fabric sample image and the standard test fabric sample image as training data of the GAN model; generating simulated test images similar to the standard test fabric sample image according to different test conditions through the generator; judging the authenticity of the simulated test image through the discriminator; the specific training process is: fixing the discriminator, training the generator, and generating simulated test images; fixing the generator, training the discriminator, and judging the authenticity of the simulated test images generated by the generator; alternately training the generator and the discriminator through the GAN loss function to generate simulated test images that are most similar to the standard test fabric sample image; after the GAN model training is completed, the potential features of the standard test fabric sample image are obtained through the intermediate layer of the generator and the high-level convolution layer of the discriminator to obtain the GAN potential features; combining the GAN potential features with the color change features to obtain a multi-dimensional feature vector; According to the multidimensional feature vector, the SVM model is used to classify the color change level of the standard test fabric sample image to analyze the influence of different factors on the color change of highly dyed polyester fabric; the classification and prediction process of the SVM model includes: inputting the multidimensional feature vector into the SVM model, and annotating the standard test fabric sample image according to the color change level; dividing the standard test fabric sample image into a training set, a validation set and a test set; selecting the RBF kernel as the kernel function, and training the SVM model and optimizing the parameters of the SVM model through cross-validation on the validation set; using the trained SVM model to classify the color change level and predict the color difference value of the standard test fabric sample image in the test set to analyze the influence of different factors on the color change of highly dyed polyester fabric; evaluating the color change level classification result of the SVM model by calculating the accuracy; and evaluating the color difference value prediction result of the SVM model by calculating the mean square error.

2. The method for analyzing color change characteristics of highly dyed polyester fabric according to claim 1, characterized in that: The preprocessing includes: image denoising, image normalization, geometric correction and color correction; wherein, the image denoising includes: using Gaussian filtering to remove noise in the test fabric sample image; the image normalization includes: converting the RGB color space of the original fabric sample image and the test fabric sample image into the Lab color space, so that the color change is more independent of the illumination change; adjusting the brightness distribution range in the image to the standard range through the L channel of the Lab image; the geometric correction includes: rotating, translating and scaling the original fabric sample image and the test fabric sample image, finding the feature point positions between multiple images to determine the relative positions of the images, and determining and cutting out the fabric area containing the feature points through an image segmentation algorithm to remove irrelevant background information; the color correction includes: converting the RGB color space of the original fabric sample image and the test fabric area image into the Lab color space, using a color equalization algorithm to adjust the brightness of different color channels to restore the color of the original fabric sample image, and converting the adjusted Lab component back to the RGB color space to obtain the standard original fabric sample image and the standard test fabric sample image.

3. The method for analyzing color change characteristics of highly dyed polyester fabric according to claim 1, characterized in that: The color difference calculation includes: calculating the color difference between the standard original fabric sample image and the standard test fabric sample image by using the color difference formula CIEDE2000; According to the color difference shown The value of sets the automatic color change level system, including: There is no perceptible color difference; Slight color difference; Moderate color difference; Severe color difference.

4. The method for analyzing color change characteristics of highly dyed polyester fabric according to claim 1, characterized in that: The color change feature includes: a first color feature, a first texture feature and a multi-spectral feature; wherein the extraction of the first color feature includes: calculating the average value of the brightness L, the red-green color a and the yellow-blue color b in the entire image area in the Lab color space to obtain the Lab mean feature; Analyze the value of The mean and variance of the values ​​were used to evaluate the degree of fabric color change under different test conditions. Features; calculate the contrast through the brightness channel L to describe the sensitivity difference of the fabric surface and obtain the contrast feature; combine the Lab mean feature, The first color feature is obtained by extracting the first texture feature by using the gray level co-occurrence matrix to extract the energy, contrast and entropy of the fabric texture in the standard original fabric sample image and the standard test fabric sample image, and obtaining the gray level co-occurrence matrix feature; the LBP values ​​of all pixels in the fabric image are counted by the local binary pattern LBP to form a feature histogram; the mean and variance of the feature histogram are calculated to obtain the LBP feature; the gray level co-occurrence matrix feature and the LBP feature are combined as the first texture feature; the multi-spectral feature is extracted by extracting the pixel values ​​of different bands from the standard original fabric sample image and the standard test fabric sample image, including: red, green, blue and near infrared bands; the average reflectivity, maximum reflectivity, minimum reflectivity and reflectivity ratio of each band are calculated to obtain the multi-spectral feature.

5. A color change characteristic analysis system for highly dyed polyester fabrics, characterized in that: include: The multispectral image acquisition module is used to acquire multispectral images of highly dyed polyester fabric samples to obtain original fabric sample images; acquire multispectral images of highly dyed polyester fabric samples after testing to capture color changes under different test conditions to obtain test fabric sample images; A multiple condition test module is used to set different test conditions to process the highly dyed polyester fabric sample, simulate the external influencing factors suffered by the highly dyed polyester fabric during dyeing and actual use, and obtain the tested highly dyed polyester fabric sample; the test conditions include: dyeing temperature conditions, color fixation time conditions, auxiliary agent conditions, exposure conditions, friction conditions, sweat stain conditions and washing conditions; the test conditions include: dyeing temperature conditions, color fixation time conditions, auxiliary agent conditions, exposure conditions, friction conditions, sweat stain conditions and washing conditions; the dyeing temperature conditions include: setting different temperatures for dyeing to achieve different dyeing conditions. The color fixing time condition includes: setting different color fixing times for testing to achieve different color fixing effects; the auxiliary agent condition includes: adding different types of auxiliary agents for dyeing; the exposure condition includes: exposing the highly dyed polyester fabric sample to natural light for testing and setting different exposure times; the friction condition includes: using a standard friction meter to perform a friction test on the highly dyed polyester fabric sample and setting different friction times; the sweat stain condition includes: using artificial sweat to test the highly dyed polyester fabric sample; the washing condition includes: using different washing methods to perform a washing test; A preprocessing module, used for preprocessing the original fabric sample image and the test fabric sample image to restore the visual color of the highly dyed polyester fabric sample to obtain a standard original fabric sample image and a standard test fabric sample image; A color difference calculation and color change level setting module, used to calculate the color difference between the standard original fabric sample image and the standard test fabric sample image, and set the color change level; A color change feature extraction module, used to extract color change features according to the standard test fabric sample image, wherein the color change features include: a first color feature, a first texture feature and a multi-spectral feature; A potential feature extraction module is used to generate a simulated test image according to different test conditions through a GAN model, and extract the GAN potential features of the generated simulated test image, including: a second texture feature, a second color feature, a structural change, an overall color distribution, and a large-scale texture change; the implementation method of the GAN model includes: inputting the standard original fabric sample image and the standard test fabric sample image as training data of the GAN model; generating a simulated test image similar to the standard test fabric sample image through a generator according to different test conditions; judging the authenticity of the simulated test image through a discriminator; the specific training process is: fixing the discriminator, training the generator, and generating a simulated test image; fixing the generator, training the discriminator, and judging the authenticity of the simulated test image generated by the generator; alternately training the generator and the discriminator through a GAN loss function to generate a simulated test image that is most similar to the standard test fabric sample image; after the GAN model is trained, the potential features of the standard test fabric sample image are obtained through the intermediate layer of the generator and the high-level convolution layer of the discriminator to obtain the GAN potential features; combining the GAN potential features with the color change features to obtain a multi-dimensional feature vector; The color change classification and color difference prediction module classifies the color change level of the standard test fabric sample image using the SVM model according to the multidimensional feature vector, so as to analyze the influence of different factors on the color change of the highly dyed polyester fabric; the classification and prediction process of the SVM model includes: inputting the multidimensional feature vector into the SVM model, and marking the standard test fabric sample image according to the color change level; dividing the standard test fabric sample image into a training set, a validation set and a test set; selecting the RBF kernel as the kernel function, and training the SVM model and optimizing the parameters of the SVM model through cross-validation on the validation set; classifying the color change level and predicting the color difference value of the standard test fabric sample image in the test set through the trained SVM model, so as to analyze the influence of different factors on the color change of the highly dyed polyester fabric; evaluating the color change level classification result of the SVM model by calculating the accuracy; and evaluating the color difference value prediction result of the SVM model by calculating the mean square error.

6. The color change characteristic analysis system of highly dyed polyester fabric according to claim 5, characterized in that: The GAN model obtains the potential features of the standard test fabric sample image through the middle layer of the generator and the high-level convolution layer of the discriminator to obtain the GAN potential features.

7. The color change characteristic analysis system of highly dyed polyester fabric according to claim 5, characterized in that: The classification and prediction process of the SVM model includes: inputting the multidimensional feature vector into the SVM model, and annotating the standard test fabric sample image according to the color change level; dividing the standard test fabric sample image into a training set and a test set; selecting the RBF kernel as the kernel function, and optimizing the parameters of the SVM model through cross-validation; using the trained SVM model to classify the color change level and predict the color difference value of the standard test fabric sample image in the test set, so as to analyze the influence of different factors on the color change of highly dyed polyester fabric; evaluating the color change level classification result of the SVM model by calculating the accuracy, precision and recall rate; and evaluating the color difference value prediction result of the SVM model by calculating the mean square error.

Images