Multi-scale texture feature extraction and intelligent discrimination algorithm and system in online quality detection of decorative paper

Abstract

The invention relates to the technical field of decorative paper quality detection, and discloses a multi-scale texture feature extraction and intelligent discrimination algorithm and system in decorative paper online quality detection.The algorithm comprises the steps that firstly, a high-definition texture image of a decorative paper sample to be detected is collected, and a three-level scale decomposition threshold value is preset; and performing three-level scale decomposition and feature extraction on the high-definition texture image to obtain a multi-scale texture image group and a multi-scale texture feature set. Then constructing a texture quality attention model to obtain a multi-scale attention weight vector set; based on the multi-scale attention weight vector set, performing weighted fusion on the multi-scale texture features to obtain a multi-scale fusion feature set; and finally, performing quality judgment on the to-be-detected decorative paper sample according to the multi-scale fusion features to obtain a quality judgment grade of the to-be-detected decorative paper sample. According to the method, the three-dimensional morphology of the reticulate pattern surface can be accurately reconstructed, the morphological parameters of the reticulate patterns are quantitatively extracted, and the printing quality and efficiency of gravure printing of the decorative paper are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of decorative paper quality detection, and more specifically, to a multi-scale texture feature extraction and intelligent discrimination algorithm and system in online quality detection of decorative paper. Background Art

[0002] With the continuous advancement of science and technology and the improvement of people's living standards, gravure printing of decorative paper has become an indispensable process in the manufacture of decorative materials such as furniture and flooring. Decorative paper gravure printing has high requirements for printing quality. How to detect the printing quality of decorative paper online and adjust the printing parameters in time has become a key issue that needs to be solved in the decorative paper printing industry.

[0003] A Chinese patent with publication number CN104034424A discloses a method for detecting and analyzing the printing quality of decorative paper. The method uses a spectrophotometer to scan the measurement and control strips on the standard sample and the current printed sheet to obtain the spectral data of each color block, further calculate the chromaticity value and dot expansion value of each color block, and evaluate the printing quality by comparing the chromaticity value and dot expansion value of the standard sample with the current printed sheet, and indicate the printing process adjustment information that needs to be taken. This patent solves the problem that the detection and control of the printing quality of decorative paper relies entirely on the experience of workers and lacks quantitative detection and control technology. However, the patent fails to achieve automatic positioning and precise quantitative analysis of printing quality problems, and subsequent optimization and adjustment of the printing process still needs to rely on manual experience.

[0004] A Chinese patent with the authorization announcement number CN113610850B discloses a method for detecting abnormal texture of decorative paper based on image processing. The method first collects the surface image of the wood grain paper to be tested, extracts the morphological features of the edge to be tested and the real edge, calculates the feature similarity between the two, and obtains multiple target real edges similar to the edge to be tested; then obtains the trend vector of each pixel point of the edge to be tested and the real edge, calculates the trend similarity between the set of trend vectors to be tested and the set of real trend vectors; finally, the texture smoothness of the edge to be tested is obtained by the feature similarity and trend similarity to determine whether the wood grain paper is abnormal. The patent uses feature similarity and trend similarity to evaluate the smoothness of edge texture, which can detect subtle edge defects and improve the accuracy of abnormality detection. However, the patent is mainly aimed at the detection of edge texture of wood grain paper, and has limited ability to evaluate the overall quality of decorative paper surface texture. In addition, the patent does not consider the fusion analysis of texture features of different scales, and the comprehensiveness and accuracy of texture quality judgment need to be further improved.

[0005] In summary, the decorative paper printing quality detection method in the existing technology lacks the extraction and fusion analysis of multi-scale texture features, making it difficult to comprehensively and accurately characterize the surface texture quality of decorative paper; the printing quality evaluation results are insufficiently interpretable, and there is a lack of reliable quantitative basis for subsequent printing parameter optimization. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art, the present invention provides a multi-scale texture feature extraction and intelligent discrimination algorithm and system for online quality detection of decorative paper, so as to realize multi-scale feature extraction, fusion and intelligent analysis of the surface texture of decorative paper, and establish a quantitative evaluation and optimization adjustment mechanism for printing quality.

[0007] To achieve the above object, the present invention provides the following technical solutions: Multi-scale texture feature extraction and intelligent discrimination algorithm for online quality inspection of decorative paper, including: Collecting a high-definition texture image of a decorative paper sample to be detected, presetting a three-level scale decomposition threshold, performing three-level scale decomposition and feature extraction on the high-definition texture image according to the preset three-level scale decomposition threshold, and obtaining a multi-scale texture image group and a multi-scale texture feature set; evaluating the effect of the three-level scale decomposition; the multi-scale texture feature set includes a first-scale texture feature, a second-scale texture feature, and a third-scale texture feature; Constructing a texture quality attention model, obtaining a multi-scale attention weight vector set according to the multi-scale texture feature set and the texture quality attention model; performing weighted fusion on the first-scale texture features, the second-scale texture features and the third-scale texture features based on the multi-scale attention weight vector set to obtain a multi-scale fusion feature set; According to the multi-scale fusion features, the quality of the decorative paper sample to be tested is judged, and the quality judgment grade of the decorative paper sample to be tested is obtained.

[0008] Furthermore, the preset three-level scale decomposition threshold includes: A scale factor is introduced, and a three-level scale decomposition threshold calculation formula is established according to the size, resolution Res and scale factor of the high-definition texture image to calculate the three-level scale decomposition threshold. The three-level scale decomposition thresholds are respectively a fine scale decomposition threshold T1, a medium scale decomposition threshold T2 and a large scale decomposition threshold T3; the size of the high-definition texture image includes an image width Wid and an image height Height.

[0009] Furthermore, the three-level scale decomposition of the high-definition texture image includes: Gaussian smoothing is performed on the high-definition texture image, and then adaptive downsampling is performed according to a fine-scale decomposition threshold T1 to generate a first-scale texture image; The first-scale texture image is Gaussian smoothed, adaptively downsampled according to the medium-scale decomposition threshold T2, and the high-frequency information of the Laplace pyramid decomposition is integrated to generate the second-scale texture image; Gaussian smoothing is performed on the second-scale texture image, adaptive downsampling is performed according to the large-scale decomposition threshold T3, and the high-frequency information of the Laplace pyramid decomposition is integrated to generate a third-scale texture image; The first scale texture image, the second scale texture image and the third scale texture image constitute a multi-scale texture image group.

[0010] Furthermore, the feature extraction of the high-definition texture image includes: Extracting first-scale texture features from the first-scale texture image; the first-scale texture features include first-scale Gabor texture features, first-scale LBP texture features and first-scale GLCM texture features; Extracting second-scale texture features from the second-scale texture image; the second-scale texture features include second-scale Gabor texture features, second-scale LBP texture features, and second-scale GLCM texture features; Extracting third-scale texture features from the third-scale texture image; the third-scale texture features include third-scale Gabor texture features, third-scale LBP texture features and third-scale GLCM texture features; The first-scale texture features, the second-scale texture features and the third-scale texture features constitute a multi-scale texture feature set.

[0011] Furthermore, the evaluation of the effect of the three-level scale decomposition includes: Calculating the structural similarity SSIM index and the scale space feature stability SSFS index of the first scale texture image, the second scale texture image, and the third scale texture image with the high-definition texture image respectively, and obtaining the SSIM index set and the SSFS index set; The SSIM index set includes three scales of SSIM indexes, and the three scales of SSIM indexes are respectively a first structural similarity SSIM index SSIM1, a second structural similarity SSIM index SSIM2 and a third structural similarity SSIM index SSIM3; The SSFS index set includes three scales of SSFS indexes, and the three scales of SSFS indexes are respectively a first scale spatial feature stability SSFS index SSFS1, a second scale spatial feature stability SSFS index SSFS2 and a third scale spatial feature stability SSFS index SSFS3; Determine whether the SSIM indexes of the three scales are all greater than the SSIM similarity threshold Ts, and whether the SSFS indexes of the three scales are all greater than the SSFS stability threshold Tf; if not, adaptively adjust the scale factor in the three-level scale decomposition threshold calculation formula until the multi-scale decomposition effect is ideal.

[0012] Further, the multi-scale attention weight vector set includes a first-scale attention weight vector, a second-scale attention weight vector, and a third-scale attention weight vector; The multi-scale attention weight vector set obtained according to the multi-scale texture feature set and the texture quality attention model includes: The first-scale Gabor texture features, the first-scale LBP texture features and the first-scale GLCM texture features are spliced ​​to form a first-scale comprehensive feature; the second-scale Gabor texture features, the second-scale LBP texture features and the second-scale GLCM texture features are spliced ​​to form a second-scale comprehensive feature; the third-scale Gabor texture features, the third-scale LBP texture features and the third-scale GLCM texture features are spliced ​​to form a third-scale comprehensive feature.

[0013] Input the first-scale comprehensive features into the texture quality attention model to obtain the first-scale attention heat map; input the second-scale comprehensive features into the texture quality attention model to obtain the second-scale attention heat map; input the third-scale comprehensive features into the texture quality attention model to obtain the third-scale attention heat map; A global average pooling operation is performed on the first-scale attention heat map, the second-scale attention heat map, and the third-scale attention heat map to obtain the first-scale attention weight vector, the second-scale attention weight vector, and the third-scale attention weight vector.

[0014] Furthermore, obtaining the multi-scale fusion features includes: According to the first-scale attention weight vector and the first-scale texture feature, a first-scale fusion feature is obtained; According to the second-scale attention weight vector and the second-scale texture feature, a second-scale fusion feature is obtained; According to the third-scale attention weight vector and the third-scale texture feature, a third-scale fusion feature is obtained; The first-scale fusion features, the second-scale fusion features, and the third-scale fusion features are concatenated to generate multi-scale fusion features.

[0015] Furthermore, obtaining the first scale fusion feature includes: Multiply the first-scale attention weight vector by the first-scale Gabor texture feature to obtain the first-scale Gabor weighted feature; Multiply the first-scale attention weight vector by the first-scale LBP texture feature to obtain the first-scale LBP weighted feature; Multiply the first-scale attention weight vector by the first-scale GLCM texture feature to obtain the first-scale GLCM weighted feature; The first-scale Gabor weighted features, the first-scale LBP weighted features and the first-scale GLCM weighted features are fused to obtain the first-scale fused features.

[0016] Furthermore, the quality determination of the decorative paper sample to be detected according to the multi-scale fusion features includes: Construct a lightweight texture feature extraction network, input the multi-scale fusion features into the lightweight texture feature extraction network, and obtain multi-scale lightweight fusion features; According to the multi-scale lightweight fusion features and the pre-built quality discrimination classifier, the quality discrimination level and the quality discrimination score corresponding to the quality discrimination level are obtained; Determine whether the quality judgment score corresponding to the quality judgment level meets the preset quality level threshold. If so, output the final quality judgment level. If not, output a warning signal.

[0017] A multi-scale texture feature extraction and intelligent discrimination system for online quality detection of decorative paper, which is used to implement the above-mentioned multi-scale texture feature extraction and intelligent discrimination algorithm for online quality detection of decorative paper, and the system comprises: Image decomposition module: used for collecting high-definition texture images of decorative paper samples to be detected, presetting three-level scale decomposition thresholds, performing three-level scale decomposition and feature extraction on the high-definition texture images according to the preset three-level scale decomposition thresholds, and obtaining a multi-scale texture image group and a multi-scale texture feature set; the multi-scale texture feature set includes first-scale texture features, second-scale texture features and third-scale texture features; Decomposition effect evaluation module: used to evaluate the effect of three-level scale decomposition; Feature fusion module: used to construct a texture quality attention model, obtain a multi-scale attention weight vector set according to the multi-scale texture feature set and the texture quality attention model; based on the multi-scale attention weight vector set, perform weighted fusion on the first-scale texture features, the second-scale texture features and the third-scale texture features to obtain a multi-scale fusion feature set; Quality judgment module: Based on the multi-scale fusion features, the quality of the decorative paper sample to be tested is judged to obtain the quality judgment grade of the decorative paper sample to be tested.

[0018] An electronic device comprises a memory, a central processing unit and a computer program stored in the memory and executable on the central processing unit. When the central processing unit executes the computer program, the multi-scale texture feature extraction and intelligent discrimination algorithm in the online quality detection of decorative paper is implemented.

[0019] A computer-readable storage medium stores a computer program, which, when executed, implements the above-mentioned multi-scale texture feature extraction and intelligent discrimination algorithm in online quality detection of decorative paper.

[0020] Compared with the prior art, the present invention has the following beneficial effects: The high-resolution industrial camera is used to collect high-definition texture images of the decorative paper surface, which ensures the integrity and accuracy of the texture details and provides a high-quality data foundation for subsequent multi-scale texture analysis. The three-level scale decomposition strategy can be used to comprehensively extract texture features at different scales, including both fine local texture details and global texture distribution laws, which enhances the representation ability of texture features. The texture quality attention mechanism is integrated, and the importance of texture features at different scales is adaptively adjusted through attention weights, highlighting the key texture areas that have a greater impact on printing quality, and improving the accuracy of quality judgment. The multi-scale texture features are weighted and fused, making full use of the complementarity of textures at different scales, forming a more comprehensive and robust texture feature set, and enhancing the reliability of quality judgment. A lightweight texture feature extraction network and quality discrimination classifier are built to achieve end-to-end online detection of decorative paper printing quality, and can quantitatively output quality judgment grades and scores, providing an intuitive and quantitative basis for quality monitoring and improvement in the production process. The entire algorithm process adopts a modular design, and each module can be flexibly combined, which is easy to expand functions and optimize performance. At the same time, the synergy of software and hardware systems is achieved to form a complete online quality detection solution for decorative paper, which can significantly improve the quality management level and production efficiency of decorative paper gravure printing. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 It is a principle flow chart of the multi-scale texture feature extraction and intelligent discrimination algorithm in the online quality detection of decorative paper in the present invention; Figure 2 A flow chart of a method for obtaining a multi-scale texture image group and a multi-scale texture feature set in a multi-scale texture feature extraction and intelligent discrimination algorithm in online quality detection of decorative paper of the present invention; Figure 3 A flow chart of a method for obtaining a multi-scale attention weight vector set in a multi-scale texture feature extraction and intelligent discrimination algorithm in online quality detection of decorative paper of the present invention; Figure 4 A flow chart of a method for obtaining multi-scale fusion features in a multi-scale texture feature extraction and intelligent discrimination algorithm in online quality detection of decorative paper of the present invention; Figure 5 It is a flow chart of a method for obtaining a first-scale fusion feature in a multi-scale texture feature extraction and intelligent discrimination algorithm in online quality detection of decorative paper of the present invention; Figure 6 A flow chart of a method for obtaining a second scale fusion feature in a multi-scale texture feature extraction and intelligent discrimination algorithm in online quality detection of decorative paper of the present invention; Figure 7 A flow chart of a method for obtaining a third-scale fusion feature in a multi-scale texture feature extraction and intelligent discrimination algorithm in online quality detection of decorative paper of the present invention; Figure 8 It is a flow chart of a method for determining the quality of a decorative paper sample to be detected in the multi-scale texture feature extraction and intelligent discrimination algorithm in the online quality detection of decorative paper of the present invention; Fig. 9 It is a functional module diagram of the multi-scale texture feature extraction and intelligent discrimination system in the online quality detection of decorative paper in the present invention. DETAILED DESCRIPTION

[0023] 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.

[0024] Example 1 See also Figure 1 As shown, this embodiment provides a multi-scale texture feature extraction and intelligent discrimination algorithm for online quality detection of decorative paper, including: Step S1000, collecting a high-definition texture image of a decorative paper sample to be detected, presetting a three-level scale decomposition threshold, performing three-level scale decomposition and feature extraction on the high-definition texture image according to the preset three-level scale decomposition threshold, and obtaining a multi-scale texture image group and a multi-scale texture feature set; evaluating the effect of the three-level scale decomposition; the multi-scale texture feature set includes a first-scale texture feature, a second-scale texture feature, and a third-scale texture feature; Furthermore, step S1000 includes: Step S1100, collecting a high-definition texture image of the decorative paper sample to be tested; Specifically, a high-resolution industrial camera is used to scan and image the surface of the decorative paper sample to obtain a high-definition texture image with a resolution of no less than 1200dpi; the high resolution of the camera ensures that the collected texture image contains rich texture detail information, providing a high-quality data basis for subsequent multi-scale decomposition and feature extraction. At the same time, the scanning imaging method can quickly obtain the complete texture image of a large area of ​​decorative paper, improving the detection efficiency. The collected high-definition texture image is preprocessed, including image denoising, image enhancement and image correction; image denoising can suppress the noise interference mixed in the imaging process, and the commonly used denoising methods include median filtering, wavelet denoising, etc. Image enhancement can improve the contrast and clarity of the image, such as histogram equalization, Retinex and other algorithms. Image correction can eliminate the geometric distortion generated during the imaging process and ensure the geometric consistency of the texture image. These preprocessing operations clear the obstacles for the accurate extraction of texture features. The processed high-definition texture image is stored in the form of a data set, and the necessary sample identification information is added, which provides a unified and standardized data source for subsequent training and testing, and improves the systematic nature of data management.

[0025] Step S1100 starts with the selection of high-standard hardware equipment, uses high-resolution industrial cameras and scanning imaging to collect the surface texture of decorative paper at ultra-high resolution, and obtains a large texture image with rich details, completeness and accuracy. And through a series of preprocessing operations, remove noise interference in the image, correct distortion, improve image quality, and finally centrally manage high-quality texture image data to prepare for subsequent texture analysis. High-quality texture images are the fundamental guarantee for texture feature extraction and have a decisive influence on the final detection performance. This step lays a solid data foundation for texture feature extraction.

[0026] Step S1200, presetting a three-level scale decomposition threshold; The preset three-level scale decomposition thresholds include: A scale factor is introduced, and a three-level scale decomposition threshold calculation formula is established according to the size, resolution Res and scale factor of the high-definition texture image to calculate the three-level scale decomposition threshold. The three-level scale decomposition thresholds are respectively a fine scale decomposition threshold T1, a medium scale decomposition threshold T2 and a large scale decomposition threshold T3; the size of the high-definition texture image includes an image width Wid and an image height Height.

[0027] The calculation formula of the three-level scale decomposition threshold is: in, Indicates the scale level, with values ​​of 1, 2, and 3; For Scaling factor, k1<k2<k3, the initial value of k1 is 1, the initial value of k2 is 4, and the initial value of k3 is 16.

[0028] Specifically, the preset three-scale decomposition threshold is to set a suitable scale range for the multi-scale decomposition of the decorative paper texture image. The three-scale corresponds to the fine scale, medium scale and large scale, respectively, which can capture the morphological characteristics of the decorative paper surface texture at different observation distances. The fine scale focuses on the subtle texture of the paper, such as fiber arrangement, dot morphology, etc.; the medium scale focuses on the mesoscopic pattern of the paper surface, such as patterns, wood grain, etc.; the large scale reflects the overall texture of the paper, such as roughness, glossiness, etc. Reasonable setting of the decomposition thresholds of the three scales is crucial for the comprehensive extraction of multi-scale texture information of decorative paper.

[0029] Considering the differences in size and resolution of the original collected high-definition texture images, it is necessary to adaptively calculate the three-level scale decomposition threshold. The threshold calculation is based on the image width Wid, height Height and resolution Res, and the scale factor is introduced. , giving flexibility to threshold calculation. The image width Wid and image height Height are in pixels, and the initial values ​​of the scale factors are k1=1, k2=4, and k3=16, which means that the original image is downsampled 4 times in width and height for the medium scale and 16 times for the large scale. This initial setting can obtain a relatively balanced decomposition effect of the three scales. However, in actual applications, the scale factors can be optimized and adjusted according to the texture characteristics of the decorative paper and the quality inspection requirements.

[0030] By downsampling the original high-definition texture image with three scale thresholds, we can obtain texture images of three scales, which reflect the changes of the texture characteristics of the paper surface at different scales. The size of the threshold determines the degree of texture information retention: the smaller the threshold, the fewer texture details are retained and the rougher the pattern; on the contrary, finer textures can be depicted. Therefore, by adjusting the scale factor, the fineness of texture decomposition can be flexibly controlled at different scales to meet the detection requirements of different types of decorative paper.

[0031] The initial value of k1 is 1, the initial value of k2 is 4, and the initial value of k3 is 16, which means that the fine scale retains all the information of the original texture image, the medium scale retains about 1 / 4 of the information, and the large scale retains about 1 / 16. This decomposition can better reflect the layered changes in the texture of the decorative paper, making it easier to extract discriminative multi-scale texture features in the subsequent process.

[0032] By presetting the three-level scale decomposition threshold, the multi-scale decomposition process can be quantified to make it more objective and repeatable. At the same time, adaptively calculating the threshold based on image size and resolution can ensure that the decomposition method is applicable to texture images of different specifications, with strong robustness and practicality. In the subsequent feature extraction and quality assessment, the texture images of these three scales will play an important role.

[0033] Step S1300, performing three-level scale decomposition and feature extraction on the high-definition texture image according to a preset three-level scale decomposition threshold, to obtain a multi-scale texture image group and a multi-scale texture feature set; Furthermore, if Figure 2 As shown, step S1300 includes: Step S1310, performing Gaussian smoothing on the high-definition texture image, and then adaptively downsampling according to the fine-scale decomposition threshold T1 to generate a first-scale texture image; and extracting first-scale texture features from the first-scale texture image; the first-scale texture features include first-scale Gabor texture features, first-scale LBP texture features, and first-scale GLCM texture features; Step S1320, performing Gaussian smoothing on the first-scale texture image, performing adaptive downsampling according to the medium-scale decomposition threshold T2, and fusing the high-frequency information of the Laplace pyramid decomposition to generate a second-scale texture image; and extracting second-scale texture features from the second-scale texture image; the second-scale texture features include second-scale Gabor texture features, second-scale LBP texture features, and second-scale GLCM texture features; Step S1330, performing Gaussian smoothing on the second-scale texture image, performing adaptive downsampling according to the large-scale decomposition threshold T3, and fusing the high-frequency information of the Laplace pyramid decomposition to generate a third-scale texture image; and extracting third-scale texture features from the third-scale texture image; the third-scale texture features include third-scale Gabor texture features, third-scale LBP texture features, and third-scale GLCM texture features; Step S1340: The first scale texture image, the second scale texture image and the third scale texture image form a multi-scale texture image group; and the first scale texture feature, the second scale texture feature and the third scale texture feature form a multi-scale texture feature set.

[0034] Specifically, the Gaussian smoothing in step S1310 is an image preprocessing technique. It uses a Gaussian function to smooth the image, which can remove noise and details in the image, thereby highlighting the main features of the image. Gaussian smoothing is equivalent to a low-pass filter that can smooth and blur the image.

[0035] After Gaussian smoothing, the image is adaptively downsampled according to the fine-scale decomposition threshold T1. The so-called adaptive downsampling refers to dynamically adjusting the sampling interval according to T1, rather than fixed sampling. When T1 is large, the sampling interval is small, and more image details are retained; when T1 is small, the sampling interval is large, and the image details are simplified. Through adaptive downsampling, a texture image of the first scale can be obtained. This image retains the main features of the original texture image, but the details are simplified. This is conducive to extracting the fine-scale features of the texture.

[0036] Steps S1320 and S1330 are similar to S1310, but the thresholds used are medium-scale decomposition threshold T2 and large-scale decomposition threshold T3, respectively. Compared with T1, the values ​​of T2 and T3 are smaller. This means that the sampling interval is larger and the image simplification is higher. The obtained second-scale and third-scale texture images only retain the rough outline of the original texture, and the details are greatly simplified. They are used to extract medium-scale and large-scale features of the texture, respectively.

[0037] In the downsampling of the second and third scales, the high-frequency information of the Laplacian pyramid decomposition is also integrated. The Laplacian pyramid is a multi-resolution image representation method. It recursively downsamples and smoothes the image to obtain a low-pass image sequence (Gaussian pyramid). Subtracting the Gaussian images of two adjacent layers can obtain a differential image containing high-frequency information, namely the Laplacian pyramid. High-frequency information corresponds to the detailed features of the texture. Integrating it into the second and third scale images can compensate for the details lost by downsampling, making the final multi-scale features more comprehensive.

[0038] The settings of T1, T2, and T3 are not directly related to the final sampling interval. They play a more important role in adjusting the granularity of texture features of different scales. When T1 is the largest, fine-scale texture features are obtained, corresponding to the tiny structures in the texture image; T2 is the second largest, medium-scale texture features are obtained; and T3 is the smallest, large-scale texture features are obtained, corresponding to large patterns in the texture image. The determination of the sampling interval also needs to consider factors such as the resolution of the original image, which can be obtained through experimental optimization.

[0039] Multi-scale analysis can characterize texture features at different granularities, capturing both texture details and the overall situation. This provides more abundant and reliable information for subsequent texture feature fusion and quality evaluation, and helps to comprehensively and accurately judge texture quality. At the same time, adaptive downsampling avoids feature loss caused by improper sampling interval selection and improves the robustness of the algorithm. Laplacian pyramid decomposition can compensate for the high-frequency details lost by downsampling, making the final multi-scale features more complete.

[0040] Gabor texture features use Gabor wavelet transform to perform multi-scale and multi-directional texture analysis on images at different scales and directions to capture the scale and direction information of image texture. Gabor wavelet is a type of wavelet function that simulates the human visual system and is sensitive to the directionality and scale changes of texture. By setting Gabor wavelet kernels of different frequencies and directions, texture information of different granularities and directions can be extracted. Gabor features can be expressed as statistics of Gabor wavelet coefficients at different scales and directions, such as mean, standard deviation, energy, etc.

[0041] LBP (Local Binary Pattern) texture feature is an effective local texture descriptor. It encodes neighborhood information into binary patterns by comparing the size relationship between a pixel and its neighboring pixels, and counts the histogram of these binary patterns as texture features. LBP features are highly robust to changes in illumination, rotation, etc., and can capture local texture information of images. By changing the radius and number of sampling points of the LBP operator, LBP features at different scales can be extracted.

[0042] GLCM (Gray-Level Co-occurrence Matrix) is a classic texture analysis method that describes the spatial correlation of image grayscale. It counts the frequency of simultaneous occurrence of different grayscale pixel pairs in an image in a certain direction and distance to form a grayscale co-occurrence matrix. By extracting second-order statistics such as energy, entropy, and contrast from GLCM, the roughness, regularity, and other characteristics of image texture can be quantified. By changing the direction and step size of GLCM, the spatial correlation of textures at different scales can be analyzed.

[0043] Combining the three classic texture analysis methods of Gabor, LBP, and GLCM, the multi-scale texture features of decorative paper can be comprehensively characterized from the three perspectives of frequency domain, spatial domain, and statistics, providing rich discriminant information for subsequent quality assessment. By extracting these features in three scale spaces respectively, the changing laws of the texture of decorative paper at different observation distances can be analyzed to achieve multi-granular texture quality evaluation. The advantage of this method is that it integrates multiple texture description methods, can comprehensively analyze the texture organization morphology of the paper surface, is sensitive to texture defects and differences, and provides a reliable feature representation for fine-grained decorative paper quality detection.

[0044] In step S1340, the texture images and texture features of the three scales are respectively summarized into a multi-scale texture image group and a multi-scale texture feature set, so as to prepare for the subsequent quality assessment. Multi-scale decomposition can simulate the process of human eyes observing paper texture at different distances. The texture image group intuitively shows the morphological changes of the paper surface pattern at different scales, and the texture feature set numerically quantifies these changes, which is convenient for the computer to make intelligent judgments.

[0045] Step S1400, evaluating the effect of the three-level scale decomposition and determining whether the multi-scale texture image group meets the feature extraction requirements; Further, step S1400 includes: Step S1410, respectively calculating the structural similarity SSIM index and the scale space feature stability SSFS index of the first scale texture image, the second scale texture image, and the third scale texture image with the high-definition texture image to obtain an SSIM index set and an SSFS index set; The SSIM index set includes three scales of SSIM indexes, and the three scales of SSIM indexes are respectively a first structural similarity SSIM index SSIM1, a second structural similarity SSIM index SSIM2 and a third structural similarity SSIM index SSIM3; wherein SSIM1 is the structural similarity SSIM index between the first scale texture image and the high-definition texture image, SSIM2 is the structural similarity SSIM index between the second scale texture image and the high-definition texture image, and SSIM3 is the structural similarity SSIM index between the third scale texture image and the high-definition texture image.

[0046] The SSFS index set includes SSFS indices of three scales, and the three scale SSFS indices are respectively the first scale space feature stability SSFS index SSFS1, the second scale space feature stability SSFS index SSFS2 and the third scale space feature stability SSFS index SSFS3; wherein SSFS1 is the scale space feature stability SSFS index of the first scale texture image and the high-definition texture image, SSFS2 is the scale space feature stability SSFS index of the second scale texture image and the high-definition texture image, and SSFS3 is the scale space feature stability SSFS index of the third scale texture image and the high-definition texture image.

[0047] The method for calculating the structural similarity SSIM index SSIM1 between the first scale texture image and the high-definition texture image comprises: in: represents the first scale texture image; Represents the original high-definition texture image; express The mean of express The mean of and It can be obtained by sliding a window over the image and calculating the average pixel value within each window; express The standard deviation of express The standard deviation of can also be calculated within the sliding window; express and The covariance of the two images reflects the structural correlation between the two images and can be calculated by sliding the window and The mean value of the corresponding pixel difference is obtained; and A small constant used to avoid the situation where the denominator is 0. It is usually taken as , ,in is the dynamic range of pixel values; Represents the first scale texture image The mean gradient amplitude of Represents the original high-definition texture image The mean value of the gradient amplitude reflects the clarity of the texture and can be obtained by calculating the gradient amplitude and averaging it using methods such as the Sobel operator. To adjust the factor, we can control the influence of gradient difference on the overall similarity and set it based on experience.

[0048] This formula introduces the gradient information of the texture image based on the original SSIM calculation of brightness, contrast and structure, forming a sigmoid function correction term. When the texture clarity of the two images is close, the sigmoid function approaches 1, which is The effect is small; when the gradient difference is large, the sigmoid function will be significantly reduced The value indicates that there is a significant difference in texture details between the two images.

[0049] Therefore, this formula comprehensively considers the similarity of images in terms of brightness, contrast, structural correlation, and texture details, and more comprehensively describes the similarity between texture images of different scales and the original image. The value range of is between 0 and 1. The closer to 1, the higher the structural similarity between the first-scale texture image and the original high-definition image. If the value of exceeds the preset similarity threshold, it can be considered that the decomposition effect of the first scale is better and can better preserve the structural information of the original texture.

[0050] This formula enhances the sensitivity of the SSIM index to changes in texture details, which helps to more finely evaluate the quality of multi-scale decomposition. At the same time, through the nonlinear transformation of the sigmoid function, SSIM is more sensitive to situations with large texture differences, and responds more gently to subtle differences, which is in line with the nonlinear characteristics of subjective visual evaluation. Therefore, the use of this formula can provide more targeted guidance for optimizing the effect of multi-scale texture decomposition.

[0051] The method for calculating the scale space feature stability SSFS index SSFS1 of the first scale texture image and the high-definition texture image comprises: in: Indicates the number of rows in the image, Indicates the number of columns of the image; Represents the first scale texture image At pixel position The gradient amplitude at Represents a high-definition texture image At pixel position The gradient amplitude at ; the gradient amplitude can be calculated using methods such as the Sobel operator; Represents pixels The weight coefficient at is used to highlight the contribution of the texture salient area; It can be obtained through texture saliency detection algorithms, such as spectral residual method; express The histogram distribution of express The histogram distribution of express and The correlation coefficient measures the similarity of their shapes.

[0052] To control the parameters, control the steepness of the sigmoid function; is the correlation coefficient threshold, and It is set by those skilled in the art based on experience or a large number of experiments.

[0053] The first term in the formula is a weighted gradient similarity that measures the consistency of the two images in terms of texture edges and details. Pixels with smaller gradient magnitudes contribute more to the similarity.

[0054] The second sigmoid function describes the correlation between the grayscale distribution of the two images. The larger the correlation coefficient, the closer the sigmoid function output is to 1, indicating that the grayscale distribution of the two images is closer and the global appearance of the texture is more similar.

[0055] along with and The difference in gradient and grayscale distribution increases. The value of will drop significantly. Therefore, The degree of approximation of the first-scale texture image to the original image in terms of local texture details and overall grayscale distribution can be comprehensively evaluated, providing a more fine-grained reference for optimizing the multi-scale decomposition strategy. Texture details and When the grayscale distribution is basically consistent and highly correlated, approaches 1. Therefore, we can set A threshold value (such as 0.85) is used as the standard for evaluating the scale decomposition effect.

[0056] This formula characterizes the feature stability of the scale space from two perspectives: local gradient information and global statistical features. It can more comprehensively guide the optimization of multi-scale decomposition parameters, obtain a multi-scale representation that is highly consistent with the original texture, and provide more reliable texture features for subsequent quality assessment.

[0057] Step S1420, determine whether the SSIM indexes of the three scales are all greater than the SSIM similarity threshold Ts, and whether the SSFS indexes of the three scales are all greater than the SSFS stability threshold Tf; if all are satisfied, it is considered that the multi-scale decomposition effect is ideal, and enter step S2000; if not, return to step S1200, and adaptively adjust the scale factor in the three-level scale decomposition threshold calculation formula until the multi-scale decomposition effect is ideal.

[0058] Specifically, the SSIM index measures the similarity between two images in terms of brightness, contrast, and structure. By introducing it into multi-scale texture analysis, the fidelity of texture images at different scales can be evaluated. The higher the SSIM index, the better the structural similarity between the texture image and the original image at that scale, and the more successfully the multi-scale decomposition retains the original texture information.

[0059] The SSFS index is an evaluation index designed specifically for multi-scale texture analysis, which is used to measure the stability of texture features in scale space. By calculating the weighted similarity of the gradient distribution of texture images at different scales and the correlation of the grayscale histogram, the consistency of texture features with scale changes is evaluated. The higher the SSFS index, the better the scale invariance of the texture features, and the more stable the texture morphology remains during the process of enlargement or reduction. This helps to enhance the scale adaptability of the subsequent discriminant model.

[0060] Step S1420 performs threshold determination on the SSIM and SSFS index sets to determine whether the effect of multi-scale decomposition is ideal. If the SSIM indexes of the three scales all exceed the given threshold Ts, and the SSFS indexes all exceed the threshold Tf, it is considered that the multi-scale texture image has a high structural similarity with the original image, the texture features have good scale stability, and the multi-scale decomposition has achieved the expected effect. Otherwise, it is necessary to return to step S1200, adjust the calculation formula of the scale decomposition threshold, and optimize the scale factor adaptively. , so that the generated multi-scale texture image group and multi-scale texture feature set can meet the needs of quality assessment.

[0061] When the SSIM indexes of the three scales are not all greater than the SSIM similarity threshold Ts, or the SSFS indexes of the three scales are not all greater than the SSFS stability threshold Tf, it is necessary to adaptively adjust the scale factor in the three-level scale decomposition threshold calculation formula until the multi-scale decomposition effect is ideal. The advantage of this adaptive adjustment mechanism is that the texture characteristics of different decorative paper samples vary greatly, and fixed scale decomposition parameters are difficult to adapt to all situations. By evaluating the decomposition effect and dynamically optimizing the decomposition strategy, the best scale space representation can be automatically searched, which will neither lose texture details due to being too coarse nor introduce redundant information due to being too fine. The optimized multi-scale decomposition can present the multi-level texture structure of decorative paper with appropriate spatial granularity, explore the quality problems hidden on the surface of the paper, and lay the foundation for accurate judgment.

[0062] For example, for decorative paper with fine patterns and complex textures, a larger scale decomposition threshold can be used to obtain a more nuanced multi-scale texture image, capture more local texture details, and avoid the texture features being too rough to accurately locate defects. For decorative paper with rough textures and simple patterns, a smaller scale decomposition threshold can meet the requirements, avoiding overly trivial texture representations that increase the computational burden.

[0063] Step S2000, constructing a texture quality attention model, obtaining a multi-scale attention weight vector set according to the multi-scale texture feature set and the texture quality attention model; performing weighted fusion on the first-scale texture features, the second-scale texture features, and the third-scale texture features based on the multi-scale attention weight vector set to obtain a multi-scale fusion feature set; Furthermore, step S2000 includes: Step S2100, constructing a texture quality attention model; Specifically, the texture quality attention model is a visual attention mechanism model based on deep learning. Its purpose is to learn the areas in the decorative paper texture image that are highly correlated with quality and assign higher weights to these areas to improve the accuracy of quality assessment. Through end-to-end training, the model automatically learns to extract discriminative texture features from a large amount of annotated data and optimizes the allocation of attention weights to achieve fine-grained evaluation of the texture quality of decorative paper.

[0064] First, a large number of decorative paper samples covering different quality levels and different texture types are collected to form the original data set. Industry experts are invited to score the texture quality of each sample and mark the key texture areas that affect the quality to obtain the quality score and pixel-level attention label. Then, data enhancement is performed on the sample images to simulate the deformation, stains, and lighting changes that may occur in the actual production and use of paper, so as to improve the generalization ability of the model. The training set and test set are divided according to a certain ratio.

[0065] Next, the network structure of the texture quality attention model is designed. With the classic convolutional neural network (CNN) as the backbone, two sub-networks are connected in parallel at the end of the network: the first one regresses the global texture quality score through the fully connected layer as the basis for quality assessment; the second one restores the feature map to the original resolution through upsampling, applies the Sigmoid activation function, and outputs a pixel-level attention heat map to characterize the influence weight of different regions on quality.

[0066] During the training phase, the sample images and the corresponding quality labels and attention labels are input into the model. The model parameters are continuously optimized through back propagation to minimize the quality regression error and attention prediction error, so that the model can learn to automatically extract texture quality-related features and assign reasonable regional weights. After multiple rounds of iterative training, the model can be used to evaluate the texture quality of decorative paper after convergence. By inputting a texture image of any size, the model can evaluate the quality score of the sample and generate the corresponding attention heat map, which intuitively presents the contribution of different texture areas to the overall quality. This evaluation result based on the attention mechanism is more transparent and explainable, which is conducive to discovering specific quality defects and has important guiding significance for the production process optimization and quality control of decorative paper.

[0067] In summary, the texture quality attention model uses a deep learning algorithm to imitate the human visual attention mechanism and adaptively focus on the areas in the texture of decorative paper that are closely related to quality. Combining texture feature extraction with attention weight learning not only improves the accuracy of quality assessment, but also enhances the interpretability of the results. The construction of this model requires the support of rich decorative paper big data and industry expert knowledge, and has a certain threshold for industrial-level applications. However, in the long run, intelligent and fine-grained texture quality assessment technology will inevitably become an important development direction in the field of decorative paper production and quality control. The texture quality attention model provides a feasible technical path to achieve this goal.

[0068] Step S2200, obtaining a multi-scale attention weight vector set according to the multi-scale texture feature set and the texture quality attention model; the multi-scale attention weight vector set includes a first-scale attention weight vector, a second-scale attention weight vector and a third-scale attention weight vector; Furthermore, if Figure 3 As shown, step S2200 includes: Step S2210, concatenating the first-scale Gabor texture feature, the first-scale LBP texture feature and the first-scale GLCM texture feature to form a first-scale comprehensive feature; Step S2220, concatenating the second-scale Gabor texture feature, the second-scale LBP texture feature, and the second-scale GLCM texture feature to form a second-scale comprehensive feature; Step S2230, concatenating the third-scale Gabor texture feature, the third-scale LBP texture feature and the third-scale GLCM texture feature to form a third-scale comprehensive feature; Step S2240, inputting the first-scale comprehensive features into the texture quality attention model to obtain a first-scale attention heat map; Step S2250, inputting the second-scale comprehensive features into the texture quality attention model to obtain a second-scale attention heat map; Step S2260, inputting the third-scale comprehensive features into the texture quality attention model to obtain a third-scale attention heat map; Step S2270, performing a global average pooling operation on the first-scale attention heat map, the second-scale attention heat map, and the third-scale attention heat map to obtain a first-scale attention weight vector, a second-scale attention weight vector, and a third-scale attention weight vector.

[0069] Specifically, in step S1300, the texture features of the decorative paper sample at three scales have been extracted to form a multi-scale texture feature set. These features depict the texture organization morphology of the paper surface from different angles and contain rich quality discrimination information. However, not all feature areas are key to quality assessment, and the importance of different areas varies. Therefore, it is necessary to further explore its inherent quality relevance on the basis of multi-scale texture features, highlight the features that have a greater impact on the discrimination process, and weaken redundant and interfering information. This is where the attention mechanism comes into play.

[0070] The texture quality attention model is an intelligent algorithm that can automatically learn the weights of feature importance. By inputting multi-scale texture features into the trained attention model, the contribution of different texture areas at each scale to quality judgment can be calculated, and the corresponding attention weight vector can be obtained. The term "attention" here borrows the concept of human visual attention and represents the selective attention and processing of different information by the visual system. In the decorative paper texture quality assessment task, the attention weight value obtained by the texture feature reflects the influence of the texture feature of the area on the assessment result.

[0071] From the perspective of information processing, the process of obtaining a set of multi-scale attention weight vectors can be understood as a process of feature selection and importance weighting. Through the attention model, a "quality assessment expert", the most discriminative key features are intelligently selected from a large number of multi-scale texture features and weights are assigned to them, so that they can play a greater role in subsequent quality judgment. This mechanism makes the decorative paper texture quality assessment process more efficient and accurate, overcoming the limitations of manual feature selection in traditional methods.

[0072] Gabor texture features, LBP texture features, and GLCM texture features describe the texture morphology from the frequency domain, spatial domain, and statistical perspectives at the same scale. The three complement each other and can more comprehensively characterize the complex and changeable decorative paper texture. By splicing the three features in the channel dimension and organically integrating them into a comprehensive feature, it is equivalent to achieving feature-level integration, which can further enhance the representation ability and quality relevance of the features. The length of the comprehensive feature is equal to the sum of the lengths of the three texture features. This fusion method is simple and effective, and is of great benefit to downstream attention learning and quality judgment.

[0073] For example, assuming that the Gabor, LBP, and GLCM texture features at the first scale are 128-dimensional, 256-dimensional, and 32-dimensional vectors, respectively, the combined first-scale features are a 416-dimensional vector. This 416-dimensional vector covers the feature information extracted by the three types of texture descriptors, and more meticulously depicts the texture of the decorative paper at this scale. The subsequent attention model will also assign different weights to the 416 feature dimensions to fully utilize the discriminative power of the comprehensive features.

[0074] Steps S2240-S2260 generate attention heat maps at the corresponding scales by inputting the comprehensive features of the three scales into the attention model respectively. The attention heat map is actually a grayscale image of the same size as the original texture image. The grayscale value of each pixel position in the image represents the importance of the texture feature at that position to the quality judgment. The higher the grayscale value (the brighter), the greater the impact of the texture in this area on the quality, and it should be given a higher weight in the evaluation process; conversely, the lower the grayscale value (the darker), the less important the texture in this area is, and the less impact it has on the evaluation results.

[0075] From a visual perspective, the attention heat map is equivalent to a distribution map of the importance of a region. It intuitively presents the quality correlation of different texture areas, making it clear at a glance. Through the heat map, we can quickly locate the key textures that have the greatest impact on the quality of the decorative paper samples, such as texture continuity, consistency, clarity, etc., providing a basis for the diagnosis of quality defects. The attention model is trained with a large amount of labeled data, so the heat map it generates can reflect the quality judgment ideas and focus of human experts. It is an intelligent way to learn expert experience from data and has a certain degree of interpretability.

[0076] It should be noted that since the attention heat map is at the pixel level, its resolution is the same as the input texture image. However, the length of the texture feature vector is much smaller than the number of image pixels, which requires the attention model to upsample and spatially map the features when generating the heat map, and expand the attention weight of the feature dimension to the pixel dimension. A common practice is to first map the feature vector to a two-dimensional matrix with the same height and width as the image through a fully connected layer, and then apply the Sigmoid function to normalize the attention weight to between 0 and 1, so as to obtain a heat map with the same size as the original image.

[0077] From a functional point of view, steps S2240-S2260 realize the mapping from multi-scale texture features to regional quality importance through the process of generating attention heat maps by comprehensive features of three scales. This importance-oriented feature enhancement method makes the texture features that were originally treated equally become discriminative, providing a clearer basis for subsequent quality assessment decisions. The visualization results of the multi-scale attention heat map also provide an intuitive and quantitative reference for quality analysis and process improvement in decorative paper production.

[0078] Step S2270 is to convert the pixel-level two-dimensional attention heat map into a one-dimensional attention weight vector with the same length as the texture feature vector, in preparation for subsequent feature weighting. The global average pooling operation is a commonly used feature compression and information aggregation method. It averages each channel of the input feature map and outputs a one-dimensional vector with the same number of channels. The optimized result is equivalent to counting the overall importance of each region in units of channels, which not only reduces the subsequent calculation amount, but also avoids the interference of regional outliers.

[0079] For example, the size of the first-scale attention heat map is 1024×1024, with a total of 416 channels. Global average pooling is to average the attention weights of 1024×1024 pixels in each channel, and finally obtain a 416-dimensional first-scale attention weight vector. Each element of the vector represents the average quality importance of the texture area on the corresponding dimension in the first-scale comprehensive feature. This vector is exactly the same length as the first-scale comprehensive feature, and can be multiplied element by element to give each texture feature dimension a corresponding weight, highlighting significant quality-related areas and weakening relatively minor areas. This weighted feature vector contains more refined quality discrimination information.

[0080] The generation process of the second and third scale attention weight vectors is similar, and the three weight vectors finally obtained together constitute a multi-scale attention weight vector set. This set comprehensively describes the quality importance distribution of the decorative paper texture area at three scales, and provides a quantitative weight reference for the subsequent multi-scale quality feature fusion. Using the attention weight vector as a bridge, the texture feature is mapped from the thousands-dimensional pixel space to the semantic space that is highly related to quality, realizing the sublimation and optimization of feature representation, which is the key to improving quality assessment performance.

[0081] In summary, step S2200 and its sub-steps use the attention model to represent the quality correlation learning of multi-scale texture features as an attention weight vector, and use global average pooling to achieve feature compression from pixel-level attention heat maps to regional-level weight vectors. This series of operations makes the originally homogenized massive texture features have discriminative differences, providing an importance-oriented feature optimization mechanism for quality assessment. Automatically learning quality association weights from texture features overcomes the limitations of traditional manual design and is a typical case of applying computer vision technology to decorative paper quality assessment. The introduction of the visual attention mechanism not only improves the objectivity and accuracy of decorative paper quality assessment, but also provides new ideas for the interpretability of assessment results. Online quality detection of decorative paper for intelligent manufacturing will surely benefit from this attention-oriented multi-scale texture analysis.

[0082] Step S2300, based on the multi-scale attention weight vector set, weighted fusion is performed on the first-scale texture feature, the second-scale texture feature and the third-scale texture feature to obtain a multi-scale fusion feature; Furthermore, if Figure 4 As shown, step S2300 includes: Step S2310, obtaining a first-scale fusion feature according to the first-scale attention weight vector and the first-scale texture feature; Furthermore, if Figure 5 As shown, step S2310 includes: Step S2311, multiplying the first-scale attention weight vector by the first-scale Gabor texture feature to obtain the first-scale Gabor weighted feature; Step S2312, multiplying the first-scale attention weight vector by the first-scale LBP texture feature to obtain the first-scale LBP weighted feature; Step S2313, multiplying the first-scale attention weight vector by the first-scale GLCM texture feature to obtain the first-scale GLCM weighted feature; Step S2314: Fusing the first-scale Gabor weighted features, the first-scale LBP weighted features, and the first-scale GLCM weighted features to obtain first-scale fused features.

[0083] Step S2320, obtaining a second-scale fusion feature according to the second-scale attention weight vector and the second-scale texture feature; Furthermore, if Figure 6 As shown, step S2320 includes: Step S2321, multiplying the second-scale attention weight vector by the second-scale Gabor texture feature to obtain a second-scale Gabor weighted feature; Step S2322, multiplying the second-scale attention weight vector by the second-scale LBP texture feature to obtain a second-scale LBP weighted feature; Step S2323, multiplying the second-scale attention weight vector by the second-scale GLCM texture feature to obtain a second-scale GLCM weighted feature; Step S2324: fuse the second-scale Gabor weighted features, the second-scale LBP weighted features, and the second-scale GLCM weighted features to obtain second-scale fused features.

[0084] Step S2330, obtaining a third-scale fusion feature according to the third-scale attention weight vector and the third-scale texture feature; Furthermore, if Figure 7 As shown, step S2330 includes: Step S2331, multiplying the third-scale attention weight vector by the third-scale Gabor texture feature to obtain a third-scale Gabor weighted feature; Step S2332, multiplying the third-scale attention weight vector by the third-scale LBP texture feature to obtain a third-scale LBP weighted feature; Step S2333, multiplying the third-scale attention weight vector by the third-scale GLCM texture feature to obtain a third-scale GLCM weighted feature; Step S2334: Fusing the third-scale Gabor weighted features, the third-scale LBP weighted features, and the third-scale GLCM weighted features to obtain third-scale fused features.

[0085] Step S2340: concatenate the first-scale fusion feature, the second-scale fusion feature, and the third-scale fusion feature to generate a multi-scale fusion feature.

[0086] Specifically, step S2300 introduces an attention mechanism to perform adaptive weighted fusion of multi-scale texture features to generate discriminative multi-scale fusion features. This step comprehensively considers the importance differences of texture features at different scales and adjusts their contribution to the final quality assessment result by assigning weights.

[0087] In step S2310, the fine-scale Gabor, LBP and GLCM texture features are weighted using the first-scale attention weight vector. Specifically, step S2311 multiplies the first-scale attention weight vector by the first-scale Gabor texture feature element by element to obtain the first-scale Gabor weighted feature. The Gabor feature here is a multi-scale and multi-directional texture descriptor extracted by Gabor wavelet transform, which is sensitive to the frequency domain characteristics of the texture. Multiplying by the attention weight is equivalent to adjusting the importance of the Gabor feature in different frequencies and directions. Steps S2312 and S2313 multiply the first-scale attention weight vector by the LBP feature and the GLCM feature in a similar manner to obtain the corresponding weighted features. The LBP feature describes the spatial pattern of the pixel neighborhood, and the GLCM feature describes the second-order statistical characteristics of the gray-level co-occurrence matrix. By assigning different weights to these features, certain texture patterns can be selectively emphasized or suppressed to improve the representation ability of the features. Step S2314 concatenates the three weighted features to form a first-scale fusion feature. This feature integrates the frequency domain, spatial domain, and statistical characteristics of texture at a fine scale, and has rich texture discrimination information.

[0088] Steps S2320 and S2330 perform weighted fusion of medium-scale and large-scale texture features using the attention weight vectors of the second scale and the third scale, respectively, in a similar manner to S2310. Among them, steps S2321-S2324 generate second-scale fusion features, and steps S2331-S2334 generate third-scale fusion features. The fusion features of these two scales focus on the performance of paper texture at medium observation distances and long distances, respectively, capturing the hierarchical changes in texture morphology.

[0089] Finally, step S2340 stitches the fusion features of the three scales together to generate the final multi-scale fusion feature. This feature covers the key information of the surface texture of the decorative paper in different scale spaces and has the ability to analyze texture differences at multiple granularities. By organically combining the fine, medium, and large-scale texture features, an information-rich and highly discriminative feature representation is constructed, laying a solid foundation for subsequent quality assessment.

[0090] The advantage of introducing the attention mechanism for multi-scale feature fusion is that it can adaptively adjust the importance of features at different scales, highlight the texture information that is beneficial to quality judgment, and suppress the influence of interference factors. Traditional multi-scale feature fusion often uses simple splicing or averaging, ignoring the importance differences between features. The attention weighted fusion quantitatively characterizes the discriminative contribution of features at each scale through the learned weight vector, making the fusion process more intelligent. The size of the weight reflects the relevance of the texture feature at this scale to the quality assessment task, providing a useful reference for feature selection. At the same time, since the attention weight is adaptively generated based on the texture image, the fusion method has a certain robustness and can handle decorative paper samples of different types and qualities.

[0091] For example, for a high-end decorative paper with a fine texture and clear texture, fine-scale texture features may contain more information that helps to distinguish quality, so the corresponding element value in the first-scale attention weight vector will be larger, highlighting the role of this part of the features in the fusion feature. For ordinary decorative paper with a rough texture and rough texture, medium-scale and large-scale features may be more discriminative, and the corresponding weight values ​​will be increased. This adaptive weighted fusion method can flexibly adjust the contribution of features at different levels according to the characteristics of the decorative paper texture, thereby improving the accuracy and generalization ability of quality assessment.

[0092] In summary, step S2300 realizes the adaptive weighted fusion of multi-scale texture features through the attention mechanism, and constructs a fusion feature that takes into account both fine and macro texture characteristics. This fusion method makes full use of the discriminant information at different observation scales and enhances the representation ability of the feature. At the same time, by quantifying the importance weights of features at each scale, the fusion process is more intelligent and flexible, and can adapt to decorative paper samples of different quality levels. The multi-scale fusion feature provides a reliable basis for subsequent quality assessment and helps to improve the accuracy and efficiency of decorative paper quality inspection. In practical applications, the calculation method of the attention weight can be optimized according to factors such as the type and quality requirements of the decorative paper to further improve the discriminative power and applicability of the fusion feature.

[0093] Step S3000 : performing quality judgment on the decorative paper sample to be detected according to the multi-scale fusion features to obtain the quality judgment grade of the decorative paper sample to be detected.

[0094] Furthermore, if Figure 8 As shown, step S3000 includes: Step S3100, constructing a lightweight texture feature extraction network, inputting multi-scale fusion features into the lightweight texture feature extraction network, and obtaining multi-scale lightweight fusion features; Specifically, the lightweight texture feature extraction network is a convolutional neural network model specially designed to extract multi-scale texture features of decorative paper. Compared with traditional deep convolutional networks, this network significantly reduces the number of model parameters and computational complexity while ensuring feature extraction performance, making it more suitable for deployment on quality inspection equipment in decorative paper production lines for real-time detection.

[0095] The construction process of the network is as follows: First, based on the classic lightweight convolutional neural network architecture (such as MobileNet, ShuffleNet, etc.), customized improvements are made according to the texture characteristics and quality inspection requirements of decorative paper. The network is mainly composed of several convolutional layers, pooling layers, and fully connected layers. The setting of the size and number of convolution kernels comprehensively considers factors such as texture scale, directionality, and computational efficiency. Secondly, a multi-scale feature fusion module is introduced in the middle layer of the network. Through operations such as cross-layer connection and feature splicing, feature maps of different receptive fields are aggregated to capture the diverse characteristics of paper surface texture at different scales. Finally, several 1×1 convolutional layers are set at the end of the network for dimensionality reduction and feature compression to obtain compact multi-scale lightweight fusion features.

[0096] The training process of the network adopts an end-to-end supervised learning method. First, the high-definition original texture image of the decorative paper is used as input, and the manually annotated quality grade label is used as the supervision signal. Then, the original image is processed through the preprocessing module to generate a multi-scale texture image group, and the corresponding multi-scale texture feature set is extracted. Next, the multi-scale texture features are input into the lightweight texture feature extraction network, and after multi-layer convolution, pooling and feature fusion operations, multi-scale lightweight fusion features with high-level semantics are extracted. Finally, the fused features are input into the quality classifier, mapped to the quality grade category through the softmax layer, and the cross entropy loss function is used as the optimization target. The network weights are updated using the back propagation algorithm to achieve end-to-end joint optimization of the feature extractor and classifier.

[0097] The training data uses a large number of decorative paper samples of different types and quality levels. The training set is expanded through data enhancement techniques (such as rotation, flipping, noise addition, etc.) to improve the generalization performance of the model. During the training process, a multi-scale texture image group is used as input and the corresponding quality grade label is used as output. The difference between the predicted quality grade and the actual quality grade is minimized through iterative optimization, so that the network learns a robust multi-scale texture feature representation and establishes a mapping relationship between features and quality grades.

[0098] The advantages of using a lightweight texture feature extraction network are: first, through customized network design and multi-scale feature fusion, the key features of the decorative paper surface texture can be extracted efficiently and accurately, providing a reliable basis for quality judgment. Secondly, the end-to-end learning method avoids the tedious process of manually designing features, and the feature extractor can automatically learn discriminative texture features from a large amount of data. Furthermore, the lightweight network structure can significantly reduce the consumption of computing resources, speed up the detection speed, and is more conducive to industrial deployment and application.

[0099] For example, for a wood grain decorative paper, multi-scale decomposition can be used to obtain texture images at different granularities, from the overall wood grain distribution to local wood grain details. These images are input into a lightweight texture feature extraction network, and the network can automatically learn the key features of the wood grain, such as the density of the texture, the arrangement direction, and the coarseness and fineness changes. Through feature fusion, the network can integrate feature information at different scales to form a more comprehensive and detailed feature description of the wood grain texture, thereby providing a more reliable basis for subsequent quality judgment.

[0100] The multi-scale lightweight fusion features extracted by the lightweight texture feature extraction network not only contain the structural information of the original texture image at different scales, but also incorporate the abstract semantic information learned by the network, which provides a more comprehensive description of the factors affecting the quality of decorative paper. These fusion features can be used as the input of the quality discrimination classifier, and the quality grade of decorative paper can be judged overall by combining various texture features, thereby improving the accuracy and reliability of quality inspection.

[0101] Step S3200, obtaining a quality judgment level and a quality judgment score corresponding to the quality judgment level according to the multi-scale lightweight fusion feature and the pre-built quality judgment classifier; Specifically, the quality discrimination classifier is an intelligent model based on machine learning, which is used to discriminate the quality grade of decorative paper according to the input texture features. Common classifier models include support vector machine (SVM), random forest, neural network, etc. SVM is selected as the discrimination classifier here, mainly based on the following considerations: SVM is a binary classification model based on statistical learning theory. It separates samples of different categories by finding the optimal classification hyperplane. It has a good theoretical foundation and extensive application practice.

[0102] SVM uses the kernel trick to transform nonlinear problems into linear problems for solution. By introducing the kernel function, it can process high-dimensional features and has good applicability in the field of texture analysis.

[0103] SVM only relies on support vectors (a small number of samples near the classification boundary), is less disturbed by other samples, has low training complexity, and is easy and fast to implement.

[0104] Compared with neural networks, SVM has less dependence on parameter tuning, stronger generalization ability, and is more suitable for small and medium sample training.

[0105] When constructing the SVM quality discrimination classifier, the multi-scale lightweight fusion features are used as input, the pre-calibrated quality grade of decorative paper is used as output, and the classifier is trained using training samples. The training samples include multi-scale texture images of several decorative papers and the corresponding manually calibrated quality grades. The multi-scale lightweight fusion feature vector of each sample image is obtained by feature extraction, and it is combined with the corresponding quality grade label to form a training data pair. The training data pair is randomly divided into a training set and a validation set. The training set is used for model training, and the validation set is used to evaluate model performance.

[0106] The training process uses a cross-validation method to divide the training set into K subsets again. Each time, K-1 subsets are selected for training, and the remaining 1 subset is used for validation. This cycle is repeated K times, and the average of the K validation results is taken as the model performance indicator. This cross-validation can effectively utilize limited training samples and reduce the risk of overfitting. The goal of model training is to find the optimal classification hyperplane parameters to minimize the classification error of the training samples while taking into account the maximum margin principle of the classification boundary.

[0107] The key parameters of SVM (such as kernel function type, penalty coefficient, kernel function parameters, etc.) are tuned through grid search and other methods to select the parameter combination with the best classification performance. The performance of the trained model is evaluated using the validation set, and indicators such as confusion matrix, precision, recall, and F1 score are calculated to comprehensively measure the accuracy and robustness of the classifier. If the performance meets expectations, it is determined as the final quality discrimination classification model; if the performance is not ideal, it is necessary to adjust the features or optimize the model until it meets the requirements.

[0108] When the trained SVM quality discrimination classifier is applied to the quality inspection of decorative paper, the multi-scale lightweight fusion features of the decorative paper sample to be inspected are used as input. After the classifier is inferred, the corresponding quality judgment grade (such as superior, qualified, defective, etc.) and quality judgment score are output. The quality judgment score is a real number between 0 and 1, which indicates the confidence or possibility that the sample belongs to this quality grade. The higher the score, the more likely the sample belongs to this grade; the lower the score, the less likely the sample belongs to this grade.

[0109] For example, suppose the quality discrimination classifier classifies decorative paper into three grades: A, B, and C, where A is the best grade, B is the qualified grade, and C is the unqualified grade. For a sample of decorative paper to be tested, its lightweight fusion feature vector is extracted through multi-scale texture analysis and input into the trained SVM classifier. The output result is: grade A 0.8, grade B 0.15, grade C 0.05. This shows that the sample is most likely to belong to grade A, with a judgment score of 0.8; the second most likely to belong to grade B, with a judgment score of 0.15; and the least likely to belong to grade C, with a judgment score of only 0.05. Therefore, it can be determined that the quality grade of the decorative paper sample is A, which is of good quality.

[0110] This quality judgment method based on multi-scale texture features and machine learning classifiers can quickly and accurately identify and score the quality grade of decorative paper, reduce the subjectivity and instability of manual evaluation, and improve the objectivity and repeatability of quality judgment. At the same time, the results of grading judgment are more detailed and quantitative, not only giving the quality grade, but also the grade confidence score, which is convenient for subsequent quality analysis and improvement. For samples with unqualified quality, the differences between their multi-scale texture features and those of standard samples can be analyzed to trace the causes of quality defects and guide the optimization of production processes.

[0111] In summary, this method makes full use of the advantages of multi-scale texture analysis and machine learning in decorative paper quality detection, forming an intelligent, efficient and accurate quality judgment system, which provides important technical support for improving the quality of decorative paper products.

[0112] Step S3300, determine whether the quality judgment score corresponding to the quality judgment level meets the preset quality level threshold. If so, output the final quality judgment level. If not, output a warning signal and trigger the fault diagnosis process.

[0113] Specifically, the quality grade threshold refers to the minimum acceptable quality judgment score for different quality grades of decorative paper set by the enterprise based on its own quality standards and customer requirements. When the quality judgment score of a sample is higher than or equal to the quality threshold of that grade, it is considered that the sample meets the quality requirements of that grade; conversely, if the judgment score is lower than the quality threshold, it means that the quality of the sample does not meet the standard and early warning and diagnosis are required.

[0114] Taking the three quality grades A, B and C in the above example as an example, it is assumed that the quality grade thresholds set by the enterprise are: 0.9 for grade A, 0.7 for grade B, and 0.6 for grade C. This means that for samples judged as grade A, their quality judgment score must reach 0.9 or above to be considered as qualified grade A products; for samples judged as grade B, their quality judgment score must be between 0.7 and 0.9; and for samples judged as grade C, if the quality judgment score is lower than 0.6, they are unqualified products and need to be eliminated.

[0115] In actual production, the quality level threshold in the enterprise quality standard database can be input or retrieved through the human-computer interaction interface, and compared with the quality judgment score output in step S3200 in real time. If the judgment score of the sample meets the quality threshold of its judgment level, the judgment result is output and the sample is classified into the corresponding quality level; if the judgment score is lower than the quality threshold, an early warning signal is output, indicating that the sample quality is unqualified and further inspection and diagnosis are required.

[0116] The output discrimination results can be presented in a visual way, such as marking samples of different quality levels with different colors on the detection interface, and displaying their judgment scores and grade labels, so that operators can intuitively understand the sample quality distribution. For unqualified samples that trigger early warnings, the system can automatically generate a quality report, record the multi-scale texture features, judgment scores, thresholds and other parameters of the samples, and mark possible causes of defects, such as uneven texture density, broken lines, abnormal colors, etc., to provide a basis for quality diagnosis.

[0117] When the warning signal is triggered, the fault diagnosis process can be automatically started. The purpose of this process is to analyze the reasons for the sample quality failure and propose corrective measures. The diagnosis process may include the following steps: Isolate and label unqualified samples to prevent them from entering subsequent processes.

[0118] High-definition images of unqualified samples are collected by industrial cameras or high-resolution scanners for fine-grained multi-scale texture analysis.

[0119] The multi-scale texture features of the unqualified samples are compared with those of the standard samples and historical unqualified samples to find abnormal points in the feature space and infer possible defect patterns.

[0120] Combined with the technological process and production parameters of decorative paper, the links and reasons for introducing defects are analyzed, such as raw material selection, pulp ratio, mesh embossing, drying and shaping, etc.

[0121] Based on the causes of defects, corrective and preventive measures are proposed, such as adjusting process parameters, replacing or maintaining equipment, improving raw material quality control, etc.

[0122] The diagnostic results and improvement measures are fed back to the production management system, and key process parameters and equipment status are monitored to prevent similar defects from happening again.

[0123] Through the fault diagnosis process, quality problems in the production of decorative paper can be quickly discovered and located, and timely countermeasures can be taken to reduce the production of defective products and improve the consistency and stability of product quality. At the same time, by accumulating characteristic data and diagnosis cases of unqualified samples, a knowledge base of decorative paper quality defects can be established to provide experience support for future quality judgment and diagnosis.

[0124] In summary, step S3300, by introducing quality grade thresholds and fault diagnosis processes, realizes the automatic determination, early warning and improvement of the quality of decorative paper, forming a complete quality control system. This method gives full play to the advantages of multi-scale texture analysis and machine learning in quality inspection, combines the actual quality standards of the enterprise, and manages product quality in a refined, digital and intelligent manner, which helps to improve the efficiency and quality level of decorative paper production and enhance the comprehensive strength of the enterprise in market competition.

[0125] Example 2 Based on Example 1, this embodiment provides a multi-scale texture feature extraction and intelligent discrimination system for online quality detection of decorative paper, such as Fig. 9 As shown, including: Image decomposition module: used for collecting high-definition texture images of decorative paper samples to be detected, presetting three-level scale decomposition thresholds, performing three-level scale decomposition and feature extraction on the high-definition texture images according to the preset three-level scale decomposition thresholds, and obtaining a multi-scale texture image group and a multi-scale texture feature set; the multi-scale texture feature set includes first-scale texture features, second-scale texture features and third-scale texture features; Decomposition effect evaluation module: used to evaluate the effect of three-level scale decomposition; Feature fusion module: used to construct a texture quality attention model, obtain a multi-scale attention weight vector set according to the multi-scale texture feature set and the texture quality attention model; based on the multi-scale attention weight vector set, perform weighted fusion on the first-scale texture features, the second-scale texture features and the third-scale texture features to obtain a multi-scale fusion feature set; Quality judgment module: Based on the multi-scale fusion features, the quality of the decorative paper sample to be tested is judged to obtain the quality judgment grade of the decorative paper sample to be tested.

[0126] In the image decomposition module, the preset three-level scale decomposition thresholds include: A scale factor is introduced, and a three-level scale decomposition threshold calculation formula is established according to the size, resolution Res and scale factor of the high-definition texture image to calculate the three-level scale decomposition threshold. The three-level scale decomposition thresholds are respectively a fine scale decomposition threshold T1, a medium scale decomposition threshold T2 and a large scale decomposition threshold T3; the size of the high-definition texture image includes an image width Wid and an image height Height.

[0127] The calculation formula of the three-level scale decomposition threshold is: in, Indicates the scale level, with values ​​of 1, 2, and 3; For The initial value of scale factor k1 is 1, the initial value of k2 is 4, and the initial value of k3 is 16.

[0128] In the image decomposition module, the three-level scale decomposition and feature extraction of the high-definition texture image includes: Step S1310, performing Gaussian smoothing on the high-definition texture image, and then adaptively downsampling according to the fine-scale decomposition threshold T1 to generate a first-scale texture image; and extracting first-scale texture features from the first-scale texture image; the first-scale texture features include first-scale Gabor texture features, first-scale LBP texture features, and first-scale GLCM texture features; Step S1320, performing Gaussian smoothing on the first-scale texture image, performing adaptive downsampling according to the medium-scale decomposition threshold T2, and fusing the high-frequency information of the Laplace pyramid decomposition to generate a second-scale texture image; and extracting second-scale texture features from the second-scale texture image; the second-scale texture features include second-scale Gabor texture features, second-scale LBP texture features, and second-scale GLCM texture features; Step S1330, performing Gaussian smoothing on the second-scale texture image, performing adaptive downsampling according to the large-scale decomposition threshold T3, and fusing the high-frequency information of the Laplace pyramid decomposition to generate a third-scale texture image; and extracting third-scale texture features from the third-scale texture image; the third-scale texture features include third-scale Gabor texture features, third-scale LBP texture features, and third-scale GLCM texture features; Step S1340: The first scale texture image, the second scale texture image and the third scale texture image form a multi-scale texture image group; and the first scale texture feature, the second scale texture feature and the third scale texture feature form a multi-scale texture feature set.

[0129] In the decomposition effect evaluation module, the evaluation of the effect of the three-level scale decomposition includes: Step S1410, respectively calculating the structural similarity SSIM index and the scale space feature stability SSFS index of the first scale texture image, the second scale texture image, and the third scale texture image with the high-definition texture image to obtain an SSIM index set and an SSFS index set; The SSIM index set includes three scales of SSIM indexes, and the three scales of SSIM indexes are respectively a first structural similarity SSIM index SSIM1, a second structural similarity SSIM index SSIM2 and a third structural similarity SSIM index SSIM3; wherein SSIM1 is the structural similarity SSIM index between the first scale texture image and the high-definition texture image, SSIM2 is the structural similarity SSIM index between the second scale texture image and the high-definition texture image, and SSIM3 is the structural similarity SSIM index between the third scale texture image and the high-definition texture image.

[0130] The SSFS index set includes SSFS indices of three scales, and the three scale SSFS indices are respectively the first scale space feature stability SSFS index SSFS1, the second scale space feature stability SSFS index SSFS2 and the third scale space feature stability SSFS index SSFS3; wherein SSFS1 is the scale space feature stability SSFS index of the first scale texture image and the high-definition texture image, SSFS2 is the scale space feature stability SSFS index of the second scale texture image and the high-definition texture image, and SSFS3 is the scale space feature stability SSFS index of the third scale texture image and the high-definition texture image.

[0131] Step S1420, determine whether the SSIM indexes of the three scales are all greater than the SSIM similarity threshold Ts, and whether the SSFS indexes of the three scales are all greater than the SSFS stability threshold Tf; if all are satisfied, it is considered that the multi-scale decomposition effect is ideal, and proceed to the subsequent steps; if not, the scale factor in the three-level scale decomposition threshold calculation formula is adaptively adjusted until the multi-scale decomposition effect is ideal.

[0132] In the feature fusion module, the multi-scale attention weight vector set includes a first-scale attention weight vector, a second-scale attention weight vector, and a third-scale attention weight vector; The multi-scale attention weight vector set obtained includes: Step S2210, concatenating the first-scale Gabor texture feature, the first-scale LBP texture feature and the first-scale GLCM texture feature to form a first-scale comprehensive feature; Step S2220, concatenating the second-scale Gabor texture feature, the second-scale LBP texture feature, and the second-scale GLCM texture feature to form a second-scale comprehensive feature; Step S2230, concatenating the third-scale Gabor texture feature, the third-scale LBP texture feature and the third-scale GLCM texture feature to form a third-scale comprehensive feature; Step S2240, inputting the first-scale comprehensive features into the texture quality attention model to obtain a first-scale attention heat map; Step S2250, inputting the second-scale comprehensive features into the texture quality attention model to obtain a second-scale attention heat map; Step S2260, inputting the third-scale comprehensive features into the texture quality attention model to obtain a third-scale attention heat map; Step S2270, performing a global average pooling operation on the first-scale attention heat map, the second-scale attention heat map, and the third-scale attention heat map to obtain a first-scale attention weight vector, a second-scale attention weight vector, and a third-scale attention weight vector.

[0133] In the feature fusion module, the multi-scale fusion features are obtained including: Step S2310, obtaining a first-scale fusion feature according to the first-scale attention weight vector and the first-scale texture feature; Step S2320, obtaining a second-scale fusion feature according to the second-scale attention weight vector and the second-scale texture feature; Step S2330, obtaining a third-scale fusion feature according to the third-scale attention weight vector and the third-scale texture feature; Step S2340: concatenate the first-scale fusion feature, the second-scale fusion feature, and the third-scale fusion feature to generate a multi-scale fusion feature.

[0134] The step S2310 includes: Step S2311, multiplying the first-scale attention weight vector by the first-scale Gabor texture feature to obtain the first-scale Gabor weighted feature; Step S2312, multiplying the first-scale attention weight vector by the first-scale LBP texture feature to obtain the first-scale LBP weighted feature; Step S2313, multiplying the first-scale attention weight vector by the first-scale GLCM texture feature to obtain the first-scale GLCM weighted feature; Step S2314: Fusing the first-scale Gabor weighted features, the first-scale LBP weighted features, and the first-scale GLCM weighted features to obtain first-scale fused features.

[0135] The step S2320 includes: Step S2321, multiplying the second-scale attention weight vector by the second-scale Gabor texture feature to obtain a second-scale Gabor weighted feature; Step S2322, multiplying the second-scale attention weight vector by the second-scale LBP texture feature to obtain a second-scale LBP weighted feature; Step S2323, multiplying the second-scale attention weight vector by the second-scale GLCM texture feature to obtain a second-scale GLCM weighted feature; Step S2324: fuse the second-scale Gabor weighted features, the second-scale LBP weighted features, and the second-scale GLCM weighted features to obtain second-scale fused features.

[0136] The step S2330 includes: Step S2331, multiplying the third-scale attention weight vector by the third-scale Gabor texture feature to obtain a third-scale Gabor weighted feature; Step S2332, multiplying the third-scale attention weight vector by the third-scale LBP texture feature to obtain a third-scale LBP weighted feature; Step S2333, multiplying the third-scale attention weight vector by the third-scale GLCM texture feature to obtain a third-scale GLCM weighted feature; Step S2334: Fusing the third-scale Gabor weighted features, the third-scale LBP weighted features, and the third-scale GLCM weighted features to obtain third-scale fused features.

[0137] In the quality determination module, the quality determination of the decorative paper sample to be detected includes: Step S3100, constructing a lightweight texture feature extraction network, inputting multi-scale fusion features into the lightweight texture feature extraction network, and obtaining multi-scale lightweight fusion features; Step S3200, obtaining a quality judgment level and a quality judgment score corresponding to the quality judgment level according to the multi-scale lightweight fusion feature and the pre-built quality judgment classifier; Step S3300, determine whether the quality judgment score corresponding to the quality judgment level meets the preset quality level threshold. If so, output the final quality judgment level. If not, output a warning signal and trigger the fault diagnosis process.

[0138] Example 3 This embodiment discloses an electronic device, which may include one or more processors and one or more memories, wherein the memories store computer-readable codes, which, when executed by the one or more processors, may execute the multi-scale texture feature extraction and intelligent discrimination algorithm in the online quality detection of decorative paper as described above.

[0139] The method or system according to the embodiment of the present application can also be implemented with the help of the architecture of an electronic device. The electronic device may include a bus, one or more CPUs, a read-only memory (ROM), a random access memory (RAM), a communication port connected to a network, an input / output component, a hard disk, etc. The storage device in the electronic device, such as a ROM or a hard disk, can store the multi-scale texture feature extraction and intelligent discrimination algorithm for online quality detection of decorative paper provided in the present application. The multi-scale texture feature extraction and intelligent discrimination algorithm in the online quality inspection of decorative paper may, for example, include: collecting a high-definition texture image of a decorative paper sample to be inspected, presetting a three-level scale decomposition threshold, performing three-level scale decomposition and feature extraction on the high-definition texture image according to the preset three-level scale decomposition threshold, and obtaining a multi-scale texture image group and a multi-scale texture feature set; evaluating the effect of the three-level scale decomposition; the multi-scale texture feature set includes first-scale texture features, second-scale texture features and third-scale texture features; constructing a texture quality attention model, and obtaining a multi-scale attention weight vector set according to the multi-scale texture feature set and the texture quality attention model; based on the multi-scale attention weight vector set, weightedly fusing the first-scale texture features, the second-scale texture features and the third-scale texture features to obtain a multi-scale fusion feature set; and performing quality judgment on the decorative paper sample to be inspected according to the multi-scale fusion features to obtain the quality judgment grade of the decorative paper sample to be inspected.

[0140] Furthermore, the electronic device may also include a user interface. Of course, the architecture disclosed in the present invention is only exemplary, and when implementing different devices, one or more components in the electronic device disclosed in the present invention may be omitted according to actual needs.

[0141] Example 4 This embodiment discloses a computer-readable storage medium, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the multi-scale texture feature extraction and intelligent discrimination algorithm in the online quality detection of decorative paper in the embodiment of the present application can be executed. The storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache). Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0142] In addition, according to the implementation of the present application, the process described in the above reference flow chart can be implemented as a computer software program. For example, the present application provides a non-transitory machine-readable storage medium, the non-transitory machine-readable storage medium stores machine-readable instructions, the machine-readable instructions can be executed by a processor to execute instructions corresponding to the method steps provided in the present application, for example: collecting a high-definition texture image of a decorative paper sample to be detected, presetting a three-level scale decomposition threshold, performing three-level scale decomposition and feature extraction on the high-definition texture image according to the preset three-level scale decomposition threshold, and obtaining a multi-scale texture image group and a multi-scale texture feature set; evaluating the effect of the three-level scale decomposition; the multi-scale texture feature set includes a first-scale texture feature, a second-scale texture feature, and a third-scale texture feature; constructing a texture quality attention model, and obtaining a multi-scale attention weight vector set according to the multi-scale texture feature set and the texture quality attention model; based on the multi-scale attention weight vector set, weighted fusion of the first-scale texture feature, the second-scale texture feature, and the third-scale texture feature is performed to obtain a multi-scale fusion feature set; according to the multi-scale fusion feature, quality judgment is performed on the decorative paper sample to be detected, and the quality judgment level of the decorative paper sample to be detected is obtained. When the computer program is executed by a central processing unit (CPU), the above functions defined in the method of the present application are performed.

[0143] The methods, systems, and devices of the present application may be implemented in many ways. For example, the methods, systems, and devices of the present application may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above order of steps for the method is for illustration only, and the steps of the method of the present application are not limited to the order specifically described above unless otherwise specified. In addition, in some embodiments, the present application may also be implemented as programs recorded in a recording medium, which include machine-readable instructions for implementing the method according to the present application. Therefore, the present application also covers recording media storing programs for executing the method according to the present application.

[0144] In addition, the parts of the above-mentioned technical solutions provided in the embodiments of the present application that are consistent with the implementation principles of the corresponding technical solutions in the prior art are not described in detail to avoid excessive redundancy.

[0145] The specific implementation modes as described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation mode of the present invention and is not intended to limit the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. Multi-scale texture feature extraction and intelligent discrimination algorithm in online quality inspection of decorative paper, characterized by: The algorithm includes: Collecting a high-definition texture image of a decorative paper sample to be detected, presetting a three-level scale decomposition threshold, performing three-level scale decomposition and feature extraction on the high-definition texture image according to the preset three-level scale decomposition threshold, and obtaining a multi-scale texture image group and a multi-scale texture feature set; evaluating the effect of the three-level scale decomposition; the multi-scale texture feature set includes a first-scale texture feature, a second-scale texture feature, and a third-scale texture feature; Constructing a texture quality attention model, obtaining a multi-scale attention weight vector set according to the multi-scale texture feature set and the texture quality attention model; performing weighted fusion on the first-scale texture features, the second-scale texture features and the third-scale texture features based on the multi-scale attention weight vector set to obtain a multi-scale fusion feature set; According to the multi-scale fusion features, the quality of the decorative paper sample to be tested is judged, and the quality judgment grade of the decorative paper sample to be tested is obtained.

2. The multi-scale texture feature extraction and intelligent discrimination algorithm for online quality detection of decorative paper according to claim 1, characterized in that: The preset three-level scale decomposition thresholds include: A scale factor is introduced, and a three-level scale decomposition threshold calculation formula is established according to the size, resolution Res and scale factor of the high-definition texture image to calculate the three-level scale decomposition threshold. The three-level scale decomposition thresholds are respectively a fine scale decomposition threshold T1, a medium scale decomposition threshold T2 and a large scale decomposition threshold T3; the size of the high-definition texture image includes an image width Wid and an image height Height.

3. The multi-scale texture feature extraction and intelligent discrimination algorithm for online quality detection of decorative paper according to claim 2, characterized in that: The three-level scale decomposition of the high-definition texture image comprises: Gaussian smoothing is performed on the high-definition texture image, and then adaptive downsampling is performed according to a fine-scale decomposition threshold T1 to generate a first-scale texture image; The first-scale texture image is Gaussian smoothed, adaptively downsampled according to the medium-scale decomposition threshold T2, and the high-frequency information of the Laplace pyramid decomposition is integrated to generate the second-scale texture image; Gaussian smoothing is performed on the second-scale texture image, adaptive downsampling is performed according to the large-scale decomposition threshold T3, and the high-frequency information of the Laplace pyramid decomposition is integrated to generate a third-scale texture image; The first scale texture image, the second scale texture image and the third scale texture image constitute a multi-scale texture image group.

4. The multi-scale texture feature extraction and intelligent discrimination algorithm for online quality inspection of decorative paper according to claim 3, characterized in that: The feature extraction of the high-definition texture image comprises: Extracting first-scale texture features from the first-scale texture image; the first-scale texture features include first-scale Gabor texture features, first-scale LBP texture features and first-scale GLCM texture features; Extracting second-scale texture features from the second-scale texture image; the second-scale texture features include second-scale Gabor texture features, second-scale LBP texture features, and second-scale GLCM texture features; Extracting third-scale texture features from the third-scale texture image; the third-scale texture features include third-scale Gabor texture features, third-scale LBP texture features and third-scale GLCM texture features; The first-scale texture features, the second-scale texture features and the third-scale texture features constitute a multi-scale texture feature set.

5. The multi-scale texture feature extraction and intelligent discrimination algorithm for online quality detection of decorative paper according to claim 3, characterized in that: The effect of evaluating the three-level scale decomposition includes: Calculating the structural similarity SSIM index and the scale space feature stability SSFS index of the first scale texture image, the second scale texture image, and the third scale texture image with the high-definition texture image respectively, and obtaining the SSIM index set and the SSFS index set; The SSIM index set includes three scales of SSIM indexes, and the three scales of SSIM indexes are respectively a first structural similarity SSIM index SSIM1, a second structural similarity SSIM index SSIM2 and a third structural similarity SSIM index SSIM3; The SSFS index set includes three scales of SSFS indexes, and the three scales of SSFS indexes are respectively a first scale spatial feature stability SSFS index SSFS1, a second scale spatial feature stability SSFS index SSFS2 and a third scale spatial feature stability SSFS index SSFS3; Determine whether the SSIM indexes of the three scales are all greater than the SSIM similarity threshold Ts, and whether the SSFS indexes of the three scales are all greater than the SSFS stability threshold Tf; if not, adaptively adjust the scale factor in the three-level scale decomposition threshold calculation formula until the multi-scale decomposition effect is ideal.

6. The multi-scale texture feature extraction and intelligent discrimination algorithm for online quality inspection of decorative paper according to claim 4, characterized in that: The multi-scale attention weight vector set includes a first-scale attention weight vector, a second-scale attention weight vector, and a third-scale attention weight vector; The multi-scale attention weight vector set obtained according to the multi-scale texture feature set and the texture quality attention model includes: The first-scale Gabor texture features, the first-scale LBP texture features and the first-scale GLCM texture features are spliced ​​to form a first-scale comprehensive feature; the second-scale Gabor texture features, the second-scale LBP texture features and the second-scale GLCM texture features are spliced ​​to form a second-scale comprehensive feature; the third-scale Gabor texture features, the third-scale LBP texture features and the third-scale GLCM texture features are spliced ​​to form a third-scale comprehensive feature; Input the first-scale comprehensive features into the texture quality attention model to obtain the first-scale attention heat map; input the second-scale comprehensive features into the texture quality attention model to obtain the second-scale attention heat map; input the third-scale comprehensive features into the texture quality attention model to obtain the third-scale attention heat map; A global average pooling operation is performed on the first-scale attention heat map, the second-scale attention heat map, and the third-scale attention heat map to obtain the first-scale attention weight vector, the second-scale attention weight vector, and the third-scale attention weight vector.

7. The multi-scale texture feature extraction and intelligent discrimination algorithm for online quality inspection of decorative paper according to claim 6, characterized in that: The multi-scale fusion feature obtained includes: According to the first-scale attention weight vector and the first-scale texture feature, a first-scale fusion feature is obtained; According to the second-scale attention weight vector and the second-scale texture feature, a second-scale fusion feature is obtained; According to the third-scale attention weight vector and the third-scale texture feature, a third-scale fusion feature is obtained; The first-scale fusion features, the second-scale fusion features, and the third-scale fusion features are concatenated to generate multi-scale fusion features.

8. The multi-scale texture feature extraction and intelligent discrimination algorithm for online quality inspection of decorative paper according to claim 7, characterized in that: The obtaining of the first scale fusion feature comprises: Multiply the first-scale attention weight vector by the first-scale Gabor texture feature to obtain the first-scale Gabor weighted feature; Multiply the first-scale attention weight vector by the first-scale LBP texture feature to obtain the first-scale LBP weighted feature; Multiply the first-scale attention weight vector by the first-scale GLCM texture feature to obtain the first-scale GLCM weighted feature; The first-scale Gabor weighted features, the first-scale LBP weighted features and the first-scale GLCM weighted features are fused to obtain the first-scale fused features.

9. The multi-scale texture feature extraction and intelligent discrimination algorithm for online quality inspection of decorative paper according to claim 8, characterized in that: The quality determination of the decorative paper sample to be detected based on the multi-scale fusion features includes: Construct a lightweight texture feature extraction network, input the multi-scale fusion features into the lightweight texture feature extraction network, and obtain multi-scale lightweight fusion features; According to the multi-scale lightweight fusion features and the pre-built quality discrimination classifier, the quality discrimination level and the quality discrimination score corresponding to the quality discrimination level are obtained; Determine whether the quality judgment score corresponding to the quality judgment level meets the preset quality level threshold. If so, output the final quality judgment level. If not, output a warning signal.

10. A system for extracting and intelligently distinguishing multi-scale texture features in online quality inspection of decorative paper, which is used to implement the algorithm for extracting and intelligently distinguishing multi-scale texture features in online quality inspection of decorative paper according to any one of claims 1 to 9, characterized in that: The system comprises: Image decomposition module: used for collecting high-definition texture images of decorative paper samples to be detected, presetting three-level scale decomposition thresholds, performing three-level scale decomposition and feature extraction on the high-definition texture images according to the preset three-level scale decomposition thresholds, and obtaining a multi-scale texture image group and a multi-scale texture feature set; the multi-scale texture feature set includes first-scale texture features, second-scale texture features and third-scale texture features; Decomposition effect evaluation module: used to evaluate the effect of three-level scale decomposition; Feature fusion module: used to construct a texture quality attention model, obtain a multi-scale attention weight vector set according to the multi-scale texture feature set and the texture quality attention model; based on the multi-scale attention weight vector set, perform weighted fusion on the first-scale texture features, the second-scale texture features and the third-scale texture features to obtain a multi-scale fusion feature set; Quality judgment module: Based on the multi-scale fusion features, the quality of the decorative paper sample to be tested is judged to obtain the quality judgment grade of the decorative paper sample to be tested.

Images