Method and system for analyzing color characteristics of silk floss knitted short-process printed fabric

Abstract

The invention relates to the technical field of image processing, in particular to a color feature analysis method and system for a silk floss knitted short-process printed fabric, and the method comprises the steps: obtaining a first multispectral image and a second high-resolution RGB image of the silk floss knitted short-process printed fabric, and obtaining a multi-modal printed fabric image through cross-modal fusion; a low-frequency color substrate and high-frequency knitting texture noise are separated through dual-tree complex wavelet transform in combination with an attention mechanism and gradient perception filtering, an image color feature matrix is extracted, and then a multivariate image color feature coefficient is obtained; comprising a color frequency domain complexity coefficient, a color uniformity coefficient, a color texture coordination coefficient, a color stability coefficient and a color saturation balance coefficient; and finally, calculating the quality coefficient of the printed fabric based on the color characteristic coefficients, and adjusting the parameters in the printing process. The quality and the production efficiency of the silk floss knitted short-process printed fabric can be effectively improved.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a method and system for analyzing color characteristics of silk-cotton knitted short-process printed fabrics. Background Art

[0002] Image processing technology has been widely used in the quality control of silk-cotton knitted short-process printed fabrics; therefore, it is of great significance to use image analysis to perform quality analysis and production control on silk-cotton knitted short-process printed fabrics; however, traditional methods have the following problems in fabric quality analysis and production control: traditional methods do not fully consider the comprehensive analysis of multispectral images and high-resolution RGB images; traditional image filtering methods do not fully consider the comprehensive analysis of double-tree complex wavelet transform, attention mechanism and gradient-aware filtering, which easily leads to insufficient efficiency in color feature extraction; traditional methods often perform analysis through a single modal image, and often analyze based on the color feature parameters of a single image, which makes it difficult to achieve a comprehensive analysis of fabric quality and working parameters, resulting in limited fabric quality and production efficiency.

[0003] Therefore, a color feature analysis method and system for silk-cotton knitted short-process printed fabrics were proposed. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for analyzing the color characteristics of silk-cotton knitted short-process printed fabrics, by obtaining a first multispectral image and a second high-resolution RGB image of the silk-cotton knitted short-process printed fabric, and obtaining a multimodal printed fabric image through cross-modal fusion; by combining a double-tree complex wavelet transform with an attention mechanism and gradient-aware filtering to separate low-frequency color base and high-frequency knitted texture noise, extracting the image color feature matrix, and then obtaining multivariate image color feature coefficients, including a color frequency domain complexity coefficient, a color uniformity coefficient, a color texture coordination coefficient, a color stability coefficient, and a color saturation balance coefficient; finally, the quality coefficient of the printed fabric is calculated based on these color feature coefficients, and the printing process parameters are adjusted. The present invention can effectively improve the quality and production efficiency of silk-cotton knitted short-process printed fabrics.

[0005] To achieve the above object, the present invention provides the following technical solutions: A method for analyzing color characteristics of silk-cotton knitted short-process printed fabrics, comprising: S1. acquiring a first multispectral image and a second high-resolution RGB image of a silk-cotton knitted short-process printed fabric, and performing cross-modal fusion to obtain a multimodal printed fabric image; S2. performing a dual-tree complex wavelet transform on the multimodal printed fabric image, combining an attention mechanism with a gradient-aware filter to separate low-frequency color base and high-frequency knitting texture noise, and obtaining an image color feature matrix of the printed fabric; S3. Analyze the image color feature matrix to obtain multivariate image color feature coefficients; the multivariate image color feature coefficients include a color frequency domain complexity coefficient, a color uniformity coefficient, a color texture coordination coefficient, a color stability coefficient, and a color saturation balance coefficient; S4. Obtaining a first printed fabric quality coefficient according to the multivariate image color characteristic coefficient, and adjusting printing process parameters.

[0006] Preferably, the acquisition process of the multimodal printed fabric image includes: using a pre-trained ResNet-18 to extract features from a first multispectral image to obtain a first multispectral feature map; using a pre-trained VGG-16 to extract features from a second high-resolution RGB image to obtain a second high-resolution RGB feature map, and based on the attention weight, performing cross-modal alignment, attention weighted fusion and upsampling on the first multispectral feature map and the second high-resolution RGB feature map, and finally generating a multimodal printed fabric image; the multimodal printed fabric image is: ; in, It is a multimodal printed fabric image; is the attention weight of the multispectral image; is the first multi-spectral characteristic graph; It is the second high-resolution RGB feature map.

[0007] Preferably, the specific process of obtaining the image color feature matrix is: The multimodal printed fabric image is decomposed into three layers through dual-tree complex wavelet transform to obtain low-frequency components and high-frequency sub-bands in six directions; the 5×5 neighborhood energy is calculated based on each high-frequency sub-band, and the attention weight of each sub-band is dynamically allocated based on the softmax function; the gradient amplitude is calculated based on the low-frequency component; and each high-frequency sub-band is corrected according to the gradient amplitude and the sub-band attention weight; finally, the inverse transformation is performed based on the corrected high-frequency sub-bands and low-frequency components to obtain the image color feature matrix.

[0008] Preferably, the color frequency domain complexity coefficient is obtained by performing Fourier transform on the LAB color space of the image color feature matrix to obtain an amplitude spectrum, and performing annular band energy statistics based on the amplitude spectrum to obtain the color frequency domain complexity coefficient; the color uniformity coefficient is obtained by dividing the image color feature matrix into multiple windows, performing multi-scale covariance analysis in the LAB color space on each window, and performing maximum eigenvalue mapping analysis; the color texture coordination coefficient is obtained by extracting the grayscale co-occurrence matrix from the L channel of the image color feature matrix, and obtaining contrast and correlation analysis; the color stability coefficient is obtained by obtaining the color difference field and edge weight matrix based on the image color feature matrix, and analyzing according to the color difference field and edge weight matrix; the color saturation balance coefficient is obtained by calculating the saturation, partitioning according to the saturation, and then analyzing according to the pixel proportion of each area.

[0009] Preferably, the first printed fabric quality coefficient is: ; in, Indicates the quality coefficient of the first printed fabric; Indicates Multivariate image color feature coefficients; Represents the exponential adjustment coefficient.

[0010] A color feature analysis system for silk-cotton knitted short-process printed fabrics, comprising: The multimodal image acquisition and fusion module is used to obtain the first multispectral image and the second high-resolution RGB image of the silk-cotton knitted short-process printed fabric, and perform cross-modal fusion to obtain a multimodal printed fabric image; A denoising and image color feature extraction module is used to perform a dual-tree complex wavelet transform on the multimodal printed fabric image, and to separate the low-frequency color base and the high-frequency knitting texture noise by combining an attention mechanism and a gradient-aware filter, so as to obtain an image color feature matrix of the printed fabric; An image color feature analysis module is used to analyze the image color feature matrix to obtain multivariate image color feature coefficients; the multivariate image color feature coefficients include a color frequency domain complexity coefficient, a color uniformity coefficient, a color texture coordination coefficient, a color stability coefficient, and a color saturation balance coefficient; The quality analysis and parameter adjustment module is used to obtain the first printing fabric quality coefficient according to the multivariate image color characteristic coefficient and adjust the printing process parameters.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention obtains a multimodal printed fabric image by cross-modal fusion of a first multispectral image and a second high-resolution RGB image; through the multimodal printed fabric image, the color characteristics of the printed fabric can be comprehensively analyzed, and then the quality analysis of the silk-cotton knitted short-process printed fabric can be comprehensively performed, thereby effectively improving the quality and production efficiency of the silk-cotton knitted short-process printed fabric.

[0012] 2. The present invention performs double-tree complex wavelet transform on multimodal printed fabric images, combines attention mechanism with gradient perception filtering, separates low-frequency color background and high-frequency knitting texture noise, and generates an image color feature matrix. The double-tree complex wavelet transform, attention mechanism and gradient perception filtering can effectively extract color features, provide effective image preprocessing for comprehensive analysis of the color features of printed fabrics, and then comprehensively analyze the quality of silk-cotton knitted short-process printed fabrics, thereby effectively improving the quality and production efficiency of silk-cotton knitted short-process printed fabrics.

[0013] 3. The present invention performs analysis based on the image color feature matrix and extracts multivariate image color feature coefficients, which include color frequency domain complexity coefficient, color uniformity coefficient, color texture coordination coefficient, color stability coefficient and color saturation balance coefficient. The first printed fabric quality coefficient is obtained according to the multivariate image color feature coefficient, which can comprehensively analyze the quality of silk-cotton knitted short-process printed fabrics, thereby effectively improving the quality and production efficiency of silk-cotton knitted short-process printed fabrics.

[0014] 4. The present invention performs quality analysis based on the first printed fabric quality coefficient, extracts the correlation matrix between the printed fabric quality coefficient and the printing process parameters, and adjusts the printing process parameters based on the correlation matrix, the target quality coefficient and the first printed fabric quality coefficient, which can effectively improve the quality and production efficiency of silk-cotton knitted short-process printed fabrics. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic flow chart of a method for analyzing color characteristics of a silk-cotton knitted short-process printed fabric provided by an embodiment of the present invention; Figure 2 A schematic diagram of a process for obtaining a first printed fabric quality coefficient provided by an embodiment of the present invention; Figure 3 A schematic structural diagram of a silk-cotton knitted short-process printed fabric color feature analysis system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] 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.

[0017] Embodiment 1 In order to effectively improve the quality and production efficiency of silk-cotton knitted short-process printed fabrics in production line A, a color feature analysis method for silk-cotton knitted short-process printed fabrics was applied. refer to Figure 1 , which is a schematic flow chart of a method for analyzing color characteristics of a silk-cotton knitted short-process printed fabric provided by an embodiment of the present invention, comprising: A method for analyzing color characteristics of silk-cotton knitted short-process printed fabrics, comprising: S1. Obtain a first multispectral image and a second high-resolution RGB image of a silk-cotton knitted short-process printed fabric, and perform cross-modal fusion to obtain a multimodal printed fabric image; further, the first multispectral image is obtained by a multispectral camera; the second high-resolution RGB image is obtained by a high-resolution camera; Furthermore, the acquisition process of the multimodal printed fabric image includes: using the pre-trained ResNet-18 to extract features from the first multispectral image to obtain a first multispectral feature map; according to the pre-trained VGG-16, the second high-resolution RGB image is extracted to obtain a second high-resolution RGB feature map, and based on the attention weight, the first multispectral feature map and the second high-resolution RGB feature map are cross-modally aligned, attention-weighted fused and up-sampled to finally generate a multimodal printed fabric image; the multimodal printed fabric image is: ; in, It is a multimodal printed fabric image; is the attention weight of the multispectral image; is the first multi-spectral characteristic graph; It is the second high-resolution RGB feature map.

[0018] S2. performing a dual-tree complex wavelet transform on the multimodal printed fabric image, combining an attention mechanism with a gradient-aware filter to separate low-frequency color base and high-frequency knitting texture noise, and obtaining an image color feature matrix of the printed fabric; This embodiment obtains a multimodal printed fabric image by cross-modal fusion of a first multispectral image and a second high-resolution RGB image; through the multimodal printed fabric image, the color characteristics of the printed fabric can be comprehensively analyzed, and then the quality analysis of the silk-cotton knitted short-process printed fabric can be comprehensively performed, thereby effectively improving the quality and production efficiency of the silk-cotton knitted short-process printed fabric.

[0019] Furthermore, the specific process of obtaining the image color feature matrix is ​​as follows: The multimodal printed fabric image is decomposed into three layers by dual-tree complex wavelet transform to obtain low-frequency components and high-frequency sub-bands in six directions; the 5×5 neighborhood energy is calculated based on each high-frequency sub-band, and the attention weight of each sub-band is dynamically allocated based on the softmax function; the gradient amplitude is calculated based on the low-frequency component; and each high-frequency sub-band is corrected according to the gradient amplitude and the sub-band attention weight; finally, the image color feature matrix is ​​obtained by inverse transformation based on the corrected high-frequency sub-bands and low-frequency components. .

[0020] This embodiment performs a double-tree complex wavelet transform on the multimodal printed fabric image, combines the attention mechanism with gradient-aware filtering, separates the low-frequency color background and the high-frequency knitting texture noise, and generates an image color feature matrix. The double-tree complex wavelet transform, the attention mechanism and the gradient-aware filtering can effectively extract the color features, provide effective image preprocessing for comprehensively analyzing the color features of the printed fabric, and then comprehensively analyze the quality of the silk-cotton knitted short-process printed fabric, thereby effectively improving the quality and production efficiency of the silk-cotton knitted short-process printed fabric.

[0021] S3. Analyze the image color feature matrix to obtain multivariate image color feature coefficients; the multivariate image color feature coefficients include a color frequency domain complexity coefficient, a color uniformity coefficient, a color texture coordination coefficient, a color stability coefficient, and a color saturation balance coefficient; Furthermore, the color frequency domain complexity coefficient is obtained by performing Fourier transform on the LAB color space of the image color feature matrix to obtain an amplitude spectrum, and performing circular band energy statistics based on the amplitude spectrum to obtain the color frequency domain complexity coefficient; Furthermore, the LAB color space includes an L channel, an a channel, and a b channel; Furthermore, the amplitude spectrum is obtained by 2D-FFT ,in Represents the spectrum, and divides the amplitude spectrum into 5 concentric rings, and calculates the amplitude spectrum energy ratio in each ring , and based on Get the color frequency domain complexity coefficient; The color frequency domain complexity coefficient is: ; in, Represents the color frequency domain complexity coefficient; Indicates a constant An exponential function with base ; The color uniformity coefficient is obtained by dividing the image color feature matrix into multiple windows, performing multi-scale covariance analysis of the LAB color space on each window, and performing maximum eigenvalue mapping analysis; Furthermore, by analyzing the image color feature matrix Divide into 16×16 windows. Calculate the covariance matrix for the a and b channels of each window ; represents the variance of channel a; represents the variance of the b channel; Represents the covariance of channel a and channel b; and performs maximum eigenvalue mapping to obtain , specifically: ; Furthermore, a color uniformity coefficient is obtained according to the maximum eigenvalue; the color uniformity coefficient is: ; in, Indicates the color uniformity coefficient; Indicates a constant An exponential function with base ; represents the maximum eigenvalue; represents the adjustment parameter; The color texture coordination coefficient is obtained by extracting the gray level co-occurrence matrix from the L channel of the image color feature matrix, and obtaining the contrast and correlation analysis in each direction; each direction includes 0°, 45°, 90° and 135°; Furthermore, from the image color feature matrix The L channel of the gray level co-occurrence matrix is ​​extracted; and the contrast and correlation features are calculated based on the gray level co-occurrence matrix; the color texture coordination coefficient is obtained based on the contrast and correlation; the color texture coordination coefficient is: ; in, Represents the color texture coordination coefficient; Indicates the correlation, reflecting the degree of dependence between pixels; Indicates contrast; The color stability coefficient is obtained based on the image color feature matrix to obtain the color difference field and the edge weight matrix, and is analyzed according to the color difference field and the edge weight matrix; Furthermore, based on the image color feature matrix The color difference field is obtained by comparing with the reference template matrix, and the color stability coefficient is obtained based on the color difference field and the edge weight matrix detected by the Sobel operator; the color stability coefficient is: ; in, represents the color stability coefficient; represents the color difference field matrix; represents the edge weight matrix; The color saturation balance coefficient is obtained by calculating the saturation, partitioning according to the saturation, and then analyzing according to the pixel ratio of each area.

[0022] Furthermore, the saturation is calculated first, and the interval distribution statistics are performed according to the saturation, which is divided into a low saturation area, a medium saturation area and a high saturation area; then the pixel proportion of each area is obtained, and the color saturation balance coefficient is obtained according to the pixel proportion of each area; the color saturation balance coefficient is: ; in, Indicates the color saturation balance coefficient; , and Represents the pixel proportions of low saturation area, medium saturation area and high saturation area respectively; This embodiment performs analysis based on the image color feature matrix and extracts multivariate image color feature coefficients, which include a color frequency domain complexity coefficient, a color uniformity coefficient, a color texture coordination coefficient, a color stability coefficient, and a color saturation balance coefficient. The first printed fabric quality coefficient is obtained based on the multivariate image color feature coefficient, and a comprehensive quality analysis of silk-cotton knitted short-process printed fabrics can be performed, thereby effectively improving the quality and production efficiency of silk-cotton knitted short-process printed fabrics.

[0023] S4. Obtaining a first printed fabric quality coefficient according to the multivariate image color characteristic coefficient, and adjusting printing process parameters.

[0024] Furthermore, the first printed fabric quality coefficient is: ; in, Indicates the quality coefficient of the first printed fabric; Indicates Multivariate image color feature coefficients; Represents the exponential adjustment coefficient.

[0025] Further, refer to Figure 2 A schematic diagram of a process for obtaining a first printed fabric quality coefficient provided by an embodiment of the present invention; The adjustment method of the printing process parameters is: based on the machine learning model, the correlation analysis is performed on the first printed fabric quality coefficient and the silk cotton knitted short-process printed fabric printing process parameters to obtain a correlation matrix, and the printing process parameters are adjusted based on the correlation matrix and the target quality coefficient to obtain the adjusted printing process parameters; the adjusted printing process parameters are: ; in, Indicates Improved printing process parameters; Indicates Correlation coefficient between the first printing fabric process parameter and the first printing fabric quality coefficient; represents the incidence matrix; Indicates the number of parameters of the printing process to be adjusted; represents the target quality coefficient; Indicates the quality coefficient of the first printed fabric; Indicates Initial printing process parameters.

[0026] Furthermore, the printing process parameters include printing speed and printing pressure.

[0027] This embodiment performs quality analysis based on the first printed fabric quality coefficient, extracts the correlation matrix between the printed fabric quality coefficient and the printing process parameters, and adjusts the printing process parameters based on the correlation matrix, the target quality coefficient and the first printed fabric quality coefficient, which can effectively improve the quality and production efficiency of silk-cotton knitted short-process printed fabrics.

[0028] This embodiment obtains the first multispectral image and the second high-resolution RGB image of the silk-cotton knitted short-process printed fabric, and obtains a multimodal printed fabric image through cross-modal fusion; separates the low-frequency color base and the high-frequency knitted texture noise through double-tree complex wavelet transform combined with attention mechanism and gradient perception filtering, extracts the image color feature matrix, and then obtains the multivariate image color feature coefficients, including color frequency domain complexity coefficient, color uniformity coefficient, color texture coordination coefficient, color stability coefficient and color saturation balance coefficient; finally, the quality coefficient of the printed fabric is calculated based on these color feature coefficients, and the process parameters are adjusted. The present invention can effectively improve the quality and production efficiency of silk-cotton knitted short-process printed fabrics.

[0029] In order to verify the effectiveness of a color feature analysis method for silk-cotton knitted short-process printed fabrics provided in this embodiment, different methods are applied to the silk-cotton knitted short-process printed fabrics of production line A, including method 1, method 2, method 3, method 4 and method 5; the quality pass rate of production line A under different methods is compared for verification; method 1 is a color feature analysis method for silk-cotton knitted short-process printed fabrics provided in this embodiment; method 2 is based on method 1 without considering multimodal images; method 2 is based on method 1 without considering noise reduction and image color feature extraction analysis; method 4 is based on method 1 without considering the first printed fabric quality coefficient analysis; method 5 is based on method 1 without considering the adjustment of printing process parameters; the specific results are shown in Table 1; Table 1 Comparison of the quality pass rate of silk-cotton knitted short-process printed fabrics on production line A using different methods

[0030] It can be seen from Table 1 that the color feature analysis method of the silk-cotton knitted short-process printed fabric proposed in this embodiment has certain effectiveness and can effectively improve the quality and production efficiency of the silk-cotton knitted short-process printed fabric.

[0031] Embodiment 2 In order to effectively improve the quality and production efficiency of silk-cotton knitted short-process printed fabrics in production line B, a color feature analysis method for silk-cotton knitted short-process printed fabrics was applied. refer to Figure 1 , which is a schematic flow chart of a method for analyzing color characteristics of a silk-cotton knitted short-process printed fabric provided by an embodiment of the present invention, comprising: A method for analyzing color characteristics of silk-cotton knitted short-process printed fabrics, comprising: The multimodal image acquisition and fusion module is used to obtain a first multispectral image and a second high-resolution RGB image of the silk-cotton knitted short-process printed fabric, and perform cross-modal fusion to obtain a multimodal printed fabric image; further, the first multispectral image is obtained by a multispectral camera; the second high-resolution RGB image is obtained by a high-resolution camera; Furthermore, the acquisition process of the multimodal printed fabric image includes: using the pre-trained ResNet-18 to extract features from the first multispectral image to obtain a first multispectral feature map; according to the pre-trained VGG-16, the second high-resolution RGB image is extracted to obtain a second high-resolution RGB feature map, and based on the attention weight, the first multispectral feature map and the second high-resolution RGB feature map are cross-modally aligned, attention-weighted fused and up-sampled to finally generate a multimodal printed fabric image; the multimodal printed fabric image is: ; in, It is a multimodal printed fabric image; is the attention weight of the multispectral image; is the first multi-spectral characteristic graph; It is the second high-resolution RGB feature map.

[0032] A denoising and image color feature extraction module is used to perform a dual-tree complex wavelet transform on the multimodal printed fabric image, and to separate the low-frequency color base and the high-frequency knitting texture noise by combining an attention mechanism and a gradient-aware filter, so as to obtain an image color feature matrix of the printed fabric; Furthermore, the specific process of obtaining the image color feature matrix is ​​as follows: The multimodal printed fabric image is decomposed into three layers by dual-tree complex wavelet transform to obtain low-frequency components and high-frequency sub-bands in six directions; the 5×5 neighborhood energy is calculated based on each high-frequency sub-band, and the attention weight of each sub-band is dynamically allocated based on the softmax function; the gradient amplitude is calculated based on the low-frequency component; and each high-frequency sub-band is corrected according to the gradient amplitude and the sub-band attention weight; finally, the image color feature matrix is ​​obtained by inverse transformation based on the corrected high-frequency sub-bands and low-frequency components. .

[0033] An image color feature analysis module is used to analyze the image color feature matrix to obtain multivariate image color feature coefficients; the multivariate image color feature coefficients include a color frequency domain complexity coefficient, a color uniformity coefficient, a color texture coordination coefficient, a color stability coefficient, and a color saturation balance coefficient; Furthermore, the color frequency domain complexity coefficient is obtained by performing Fourier transform on the LAB color space of the image color feature matrix to obtain an amplitude spectrum, and performing circular band energy statistics based on the amplitude spectrum to obtain the color frequency domain complexity coefficient; Furthermore, the LAB color space includes an L channel, an a channel, and a b channel; Furthermore, the amplitude spectrum is obtained by 2D-FFT ,in Represents the spectrum, and divides the amplitude spectrum into 5 concentric rings, and calculates the amplitude spectrum energy ratio in each ring , and based on Get the color frequency domain complexity coefficient; The color frequency domain complexity coefficient is: ; in, Represents the color frequency domain complexity coefficient; Indicates a constant An exponential function with base ; The color uniformity coefficient is obtained by dividing the image color feature matrix into multiple windows, performing multi-scale covariance analysis of the LAB color space on each window, and performing maximum eigenvalue mapping analysis. ; Furthermore, by analyzing the image color feature matrix Divide into 16×16 windows. Calculate the covariance matrix for the a and b channels of each window ; represents the variance of channel a; represents the variance of the b channel; Represents the covariance of channel a and channel b; and performs maximum eigenvalue mapping to obtain , specifically: ; Furthermore, a color uniformity coefficient is obtained according to the maximum eigenvalue; the color uniformity coefficient is: ; in, Indicates the color uniformity coefficient; Indicates a constant An exponential function with base ; represents the maximum eigenvalue; represents the adjustment parameter; The color texture coordination coefficient is obtained by extracting the gray level co-occurrence matrix from the L channel of the image color feature matrix, and obtaining the contrast and correlation analysis in each direction; each direction includes 0°, 45°, 90° and 135°; Furthermore, from the image color feature matrix The L channel of the gray level co-occurrence matrix is ​​extracted; and the contrast and correlation features are calculated based on the gray level co-occurrence matrix; the color texture coordination coefficient is obtained based on the contrast and correlation; the color texture coordination coefficient is: ; in, Represents the color texture coordination coefficient; Indicates the correlation, reflecting the degree of dependence between pixels; Indicates contrast; The color stability coefficient is obtained based on the image color feature matrix to obtain the color difference field and the edge weight matrix, and is analyzed according to the color difference field and the edge weight matrix; Furthermore, based on the image color feature matrix The color difference field is obtained by comparing with the reference template matrix, and the color stability coefficient is obtained based on the color difference field and the edge weight matrix detected by the Sobel operator; the color stability coefficient is: ; in, represents the color stability coefficient; represents the color difference field matrix; represents the edge weight matrix; The color saturation balance coefficient is obtained by calculating the saturation, partitioning according to the saturation, and then analyzing according to the pixel ratio of each area.

[0034] Furthermore, the saturation is calculated first, and the interval distribution statistics are performed according to the saturation, which is divided into a low saturation area, a medium saturation area and a high saturation area; then the pixel proportion of each area is obtained, and the color saturation balance coefficient is obtained according to the pixel proportion of each area; the color saturation balance coefficient is: ; in, Indicates the color saturation balance coefficient; , and Represents the pixel proportions of low saturation area, medium saturation area and high saturation area respectively; The quality analysis and parameter adjustment module is used to obtain the first printing fabric quality coefficient according to the multivariate image color characteristic coefficient and adjust the printing process parameters.

[0035] Furthermore, the first printed fabric quality coefficient is: ; in, Indicates the quality coefficient of the first printed fabric; Indicates Multivariate image color feature coefficients; Represents the exponential adjustment coefficient.

[0036] Further, refer to Figure 2 A schematic diagram of a process for obtaining a first printed fabric quality coefficient provided by an embodiment of the present invention; The adjustment method of the printing process parameters is: based on the machine learning model, the correlation analysis is performed on the first printed fabric quality coefficient and the silk cotton knitted short-process printed fabric printing process parameters to obtain a correlation matrix, and the printing process parameters are adjusted based on the correlation matrix and the target quality coefficient to obtain the adjusted printing process parameters; the adjusted printing process parameters are: ; in, Indicates Improved printing process parameters; Indicates Correlation coefficient between the first printing fabric process parameter and the first printing fabric quality coefficient; represents the incidence matrix; Indicates the number of parameters of the printing process to be adjusted; represents the target quality coefficient; Indicates the quality coefficient of the first printed fabric; Indicates Initial printing process parameters.

[0037] Furthermore, the printing process parameters include printing speed and printing pressure. In order to verify the effectiveness of the silk-cotton knitted short-process printed fabric color feature analysis system provided in this embodiment, different systems are applied to the silk-cotton knitted short-process printed fabric of production line B, including system 1, system 2, system 3, system 4 and system 5; the quality pass rate of production line B under different system applications is compared for verification; system 1 is a silk-cotton knitted short-process printed fabric color feature analysis system provided in this embodiment; system 2 is based on system 1 without considering multimodal images; system 2 is based on system 1 without considering noise reduction and image color feature extraction analysis; system 4 is based on system 1 without considering the first printed fabric quality coefficient analysis; system 5 is based on system 1 without considering the adjustment of printing process parameters; the specific results are shown in Table 2; Table 2 Comparison of the quality pass rates of silk-cotton knitted short-process printed fabrics on production line B by different systems

[0038] It can be seen from Table 2 that the color feature analysis system for silk-cotton knitted short-process printed fabrics proposed in this embodiment has certain effectiveness and can effectively improve the quality and production efficiency of silk-cotton knitted short-process printed fabrics.

[0039] 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 color characteristics of silk-cotton knitted short-process printed fabrics, characterized in that: include: S1. Obtaining a first multispectral image and a second high-resolution RGB image of a silk-cotton knitted short-process printed fabric, and performing cross-modal fusion to obtain a multimodal printed fabric image; S2. performing a dual-tree complex wavelet transform on the multimodal printed fabric image, combining an attention mechanism with a gradient-aware filter to separate low-frequency color base and high-frequency knitting texture noise, and obtaining an image color feature matrix of the printed fabric; S3. Analyze the image color feature matrix to obtain multivariate image color feature coefficients; The multivariate image color feature coefficients include color frequency domain complexity coefficient, color uniformity coefficient, color texture coordination coefficient, color stability coefficient and color saturation balance coefficient; S4. Obtaining a first printed fabric quality coefficient according to the multivariate image color characteristic coefficient, and adjusting printing process parameters.

2. The method for analyzing color characteristics of silk-cotton knitted short-process printed fabrics according to claim 1, characterized in that: The acquisition process of the multimodal printed fabric image includes: using a pre-trained ResNet-18 to extract features from a first multispectral image to obtain a first multispectral feature map; using a pre-trained VGG-16 to extract features from a second high-resolution RGB image to obtain a second high-resolution RGB feature map, and based on the attention weight, performing cross-modal alignment, attention weighted fusion and upsampling on the first multispectral feature map and the second high-resolution RGB feature map to finally generate a multimodal printed fabric image.

3. The method for analyzing color characteristics of silk-cotton knitted short-process printed fabrics according to claim 2, characterized in that: The multimodal printed fabric image is: ； in, It is a multimodal printed fabric image; is the attention weight of the multispectral image; is the first multi-spectral characteristic graph; It is the second high-resolution RGB feature map.

4. The method for analyzing color characteristics of silk-cotton knitted short-process printed fabrics according to claim 3, characterized in that: The specific acquisition process of the image color feature matrix is ​​as follows: The multimodal printed fabric image is decomposed into three layers by dual-tree complex wavelet transform to obtain low-frequency components and high-frequency sub-bands in six directions; 5×5 neighborhood energy is calculated based on each high-frequency sub-band, and attention weights of each sub-band are dynamically allocated based on a softmax function; Calculate the gradient magnitude based on the low-frequency component; Each high-frequency sub-band is corrected according to the gradient amplitude and the sub-band attention weight; finally, an inverse transformation is performed based on the corrected high-frequency sub-band and low-frequency components to obtain the image color feature matrix.

5. The method for analyzing color characteristics of silk-cotton knitted short-process printed fabrics according to claim 4, characterized in that: The color frequency domain complexity coefficient is obtained by performing Fourier transform on the LAB color space of the image color feature matrix to obtain an amplitude spectrum, and performing ring band energy statistics based on the amplitude spectrum to obtain the color frequency domain complexity coefficient; the color uniformity coefficient is obtained by dividing the image color feature matrix into multiple windows, performing multi-scale covariance analysis of the LAB color space on each window, and performing maximum eigenvalue mapping analysis; The color texture coordination coefficient is obtained by extracting the gray level co-occurrence matrix from the L channel of the image color feature matrix, and obtaining the contrast and correlation analysis; The color stability coefficient is obtained by obtaining the color difference field and the edge weight matrix based on the image color feature matrix, and is analyzed according to the color difference field and the edge weight matrix; the color saturation balance coefficient is obtained by calculating the saturation, partitioning according to the saturation, and then analyzing according to the pixel proportion of each area.

6. The method for analyzing color characteristics of silk-cotton knitted short-process printed fabrics according to claim 5, characterized in that: The first printed fabric quality coefficient is: ； in, Indicates the quality coefficient of the first printed fabric; Indicates Multivariate image color feature coefficients; Represents the exponential adjustment coefficient.

7. A color feature analysis system for silk-cotton knitted short-process printed fabrics, characterized in that: include: The multimodal image acquisition and fusion module is used to obtain the first multispectral image and the second high-resolution RGB image of the silk-cotton knitted short-process printed fabric, and perform cross-modal fusion to obtain a multimodal printed fabric image; A denoising and image color feature extraction module is used to perform a dual-tree complex wavelet transform on the multimodal printed fabric image, and to separate the low-frequency color base and the high-frequency knitting texture noise by combining an attention mechanism and a gradient-aware filter, so as to obtain an image color feature matrix of the printed fabric; An image color feature analysis module, used to analyze the image color feature matrix to obtain multivariate image color feature coefficients; The multivariate image color feature coefficients include color frequency domain complexity coefficient, color uniformity coefficient, color texture coordination coefficient, color stability coefficient and color saturation balance coefficient; The quality analysis and parameter adjustment module is used to obtain the first printing fabric quality coefficient according to the multivariate image color characteristic coefficient and adjust the printing process parameters.

Images