A method and system for intelligent color separation of lining fabric patterns based on clustering and stratification

By using spectral clustering algorithm and color structure recognition model in the intelligent color separation of cloth flower patterns, combined with color psychology, the shortcomings of traditional clustering algorithms in handling complex data distribution are solved, and efficient and flexible color combination optimization is achieved. Improve the interpretability and market acceptance of the analysis results.

CN118537425BActive Publication Date: 2024-11-08QIDONG GAOYANG MECHANICAL ELECTRICAL MFG CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410622223.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-05-20
Publication Date
2024-11-08
Estimated Expiration
2044-05-20

AI Technical Summary

Technical Problem

Traditional clustering algorithms are difficult to effectively process complex or non-spherical data distributions, especially in multi-dimensional data analysis. Common algorithms cannot identify complex patterns and relationships, and lack flexibility, making it difficult to adapt to different types of data and analysis needs, and analyze results. Poor interpretability.

Method used

Using the intelligent color separation method of lining cloth flower shape based on clustering layering, the lining cloth image is acquired and preprocessed, feature data is extracted, spectral clustering algorithm and color structure recognition model are used, and similarity thresholds are dynamically set in combination with color psychology. Feature area clustering and color combination optimization.

Benefits of technology

It realizes effective processing of complex data distribution, improves the accuracy and interpretability of clustering results, is suitable for various types of data, provides scientific and market-oriented color design decision support, reduces trial and error costs, and increases The market competitiveness of the product.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118537425B_ABST
    Figure CN118537425B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for intelligent color separation of lining fabric patterns based on clustering and stratification, which is used in the field of data processing. The method comprises the following steps: obtaining an lining fabric image to be detected, preprocessing the lining fabric image to be detected, obtaining an lining fabric image with enhanced image quality, and extracting lining fabric feature data from the lining fabric image with enhanced image quality; clustering the lining fabric feature data using a clustering algorithm to identify feature areas that meet a preset similarity threshold; analyzing the pattern feature data in the feature area using a color structure recognition model and extracting color combinations, while combining color psychology to extract the preference for color combinations, and obtaining the best lining fabric pattern color combination. The present invention identifies and extracts color combinations in specific patterns through a color structure recognition model, while combining the principles of color psychology to evaluate the visual and emotional effects of different color combinations.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a clustering and stratification-based lining fabric pattern intelligent color separation method and system. Background Art

[0002] Interlining pattern usually refers to the pattern design on the fabric used as lining material in shirts, skirts and other clothing. Interlining not only plays a role in strengthening the structure of clothing, but also can increase the beauty and personalization of clothing through its pattern design. In clothing design, the choice of interlining pattern has an important influence on the overall design style and feel.

[0003] In traditional printing or dyeing processes, each color usually requires a separate plate or template. Therefore, the pattern design must be separated by color to accommodate the application of printing equipment or dyes. Color separation can help manufacturers control costs, because with each additional color, the associated plate making and printing and dyeing costs will also increase. By effectively separating colors, production costs can be controlled while maintaining the beauty of the design. The color distribution of the interlining pattern needs to be carefully designed to ensure that the colors are visually balanced and attractive. Color separation helps designers plan the application of each color in detail, making the final product more visually harmonious.

[0004] Traditional clustering algorithms often have difficulty effectively handling complex or non-spherical data distributions, especially in multidimensional data analysis. Common algorithms may not be able to identify complex patterns and relationships. Many clustering algorithms lack flexibility in similarity measurement and parameter settings, making it difficult for them to adapt to different types of data and analysis needs. The results generated by many analysis methods are difficult to display intuitively and lack sufficient interpretability, making it difficult for non-professionals to understand. At the same time, traditional color design and analysis tools often ignore psychological principles and the specific preferences of market consumers.

[0005] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0006] In order to overcome the above problems, the present invention aims to propose a method and system for intelligent color separation of lining fabric patterns based on clustering and stratification, with the aim of solving the problem that traditional clustering algorithms are often difficult to effectively handle complex or non-spherical data distributions, especially in multidimensional data analysis, common algorithms may not be able to identify complex patterns and relationships, and many clustering algorithms lack flexibility in similarity measurement and parameter setting, making it difficult for them to adapt to different types of data and analysis requirements.

[0007] To this end, the specific technical solution adopted by the present invention is as follows:

[0008] According to one aspect of the present invention, a method for intelligent color separation of lining fabric patterns based on clustering and stratification is provided, and the method for intelligent color separation of lining fabric patterns comprises the following steps:

[0009] S1, obtaining an interlining cloth image to be detected, preprocessing the interlining cloth image to be detected, obtaining an interlining cloth image with enhanced image quality, and extracting interlining cloth feature data from the interlining cloth image with enhanced image quality;

[0010] S2, clustering the lining feature data using a clustering algorithm to identify feature areas that meet a preset similarity threshold;

[0011] S3. Extract the pattern feature data in the feature area, use the color structure recognition model to analyze the pattern feature data in the feature area and extract the color combination, and combine color psychology to extract the preference of color combination to obtain the best lining cloth pattern color combination.

[0012] Optionally, clustering the lining cloth feature data using a clustering algorithm to identify feature areas that meet a preset similarity threshold comprises the following steps:

[0013] S21, obtaining lining cloth feature data, and selecting a similarity measure of the lining cloth feature data;

[0014] S22, calculating similarity based on the selected lining cloth feature data and similarity measurement, and constructing a similarity matrix;

[0015] S23, constructing a Laplace matrix according to the similarity matrix, and performing eigendecomposition to obtain eigenvalues ​​and eigenvectors of the Laplace matrix;

[0016] S24, according to the eigenvalues ​​and eigenvectors of the Laplace matrix, selecting the eigenvector corresponding to the minimum eigenvalue, and clustering the eigenvectors;

[0017] S25. According to the clustering result, the clustering result is mapped back to the lining cloth image, and each cluster is marked as a feature region.

[0018] Optionally, calculating similarity based on the selected interlining feature data and similarity measurement and constructing a similarity matrix comprises the following steps:

[0019] S221, selecting a similarity measure according to the extracted lining cloth feature data;

[0020] S222, for each pair of pixels in the lining cloth feature data, using a kernel function to calculate the similarity between feature vectors;

[0021] S223, pre-configuring a similarity threshold, and retaining similarities higher than the pre-configured similarity threshold;

[0022] S224, filling similarities higher than a pre-configured similarity threshold into the similarity matrix respectively, and constructing a similarity matrix;

[0023] Among them, the formula for calculating the similarity between feature vectors using the kernel function is:

[0024]

[0025] Where W ij Represents the similarity between pixel i and pixel j;

[0026] exp() represents the natural exponential function;

[0027] x i Represents the feature vector of pixel i;

[0028] x j represents the feature vector of pixel j;

[0029] σ represents the bandwidth parameter of the Gaussian kernel function;

[0030] ||x i -x j || 2 represents the feature vector x i and the eigenvector x j The square of the Euclidean distance between them.

[0031] Optionally, constructing a Laplace matrix according to the similarity matrix and performing eigendecomposition to obtain the eigenvalues ​​and eigenvectors of the Laplace matrix includes the following steps:

[0032] S231, based on each row in the similarity matrix, calculate the sum of all elements in each row, and set the sum to the element on the corresponding diagonal of the degree matrix;

[0033] S232, constructing a Laplace matrix according to the similarity matrix and the degree matrix;

[0034] S233. Solve the eigenvalues ​​and eigenvectors of the Laplace matrix.

[0035] Optionally, according to the eigenvalues ​​and eigenvectors of the Laplace matrix, selecting the eigenvector corresponding to the minimum eigenvalue, and clustering the eigenvectors comprises the following steps:

[0036] S241, selecting an eigenvector with a minimum eigenvalue from the eigendecomposition of the Laplace matrix;

[0037] S242, performing standardization processing on the selected feature vector;

[0038] S243, clustering the standardized feature vectors;

[0039] S244, analyzing the clustering results, checking the quality of each cluster, and judging whether the division of feature regions is reasonable;

[0040] S245. Map each cluster back to the lining cloth image according to the judgment result, wherein each cluster corresponds to each feature area.

[0041] Optionally, extracting pattern feature data in the feature area, analyzing the pattern feature data in the feature area using a color structure recognition model and extracting color combinations, and extracting color combination preferences in combination with color psychology to obtain the best lining pattern color combination includes the following steps:

[0042] S31, acquiring flower pattern feature data from the feature area;

[0043] S32, pre-configuring the framework of the color structure recognition model and matching the dimension of the flower pattern feature data;

[0044] S33, inputting the flower pattern feature data into the color structure recognition model for training, and clustering the feature parameters on adjacent neurons through self-organizing features;

[0045] S34, analyzing the output of the color structure recognition model to identify the color combination;

[0046] S35. Analyze and identify color combinations based on the principles of color psychology;

[0047] S36, combining the color combination obtained from the color structure recognition model with the color combination analyzed and recognized based on the color psychology principle, and determining the best lining fabric pattern color combination based on the combination result.

[0048] Optionally, inputting the flower pattern feature data into the color structure recognition model for training, and aggregating the feature parameters on adjacent neurons through self-organizing features comprises the following steps:

[0049] S331, determining the size and shape of the color structure recognition model, and setting the initial weights of the neurons in the competition layer;

[0050] S332, normalizing the flower pattern feature data;

[0051] S333, for each flower pattern feature data, calculating the Euclidean distance between the flower pattern feature data and the neurons in the competition layer, and determining the winning neuron;

[0052] S334, updating the weights of the winning neuron and the neurons in the neighborhood of the winning neuron according to the learning rate and the neighborhood function;

[0053] S335, repeating the steps of S333-S334, performing multiple iterations on all the flower pattern feature data, until the color structure recognition model completes a predetermined number of iterations.

[0054] Optionally, analyzing the output of the color structure recognition model to recognize color combinations comprises the following steps:

[0055] S341, after the training is completed, extracting the weight vector of each neuron from the competition layer of the color structure recognition model as output data;

[0056] S342, analyzing the extracted output data to identify different color features and clustering structures between the color features;

[0057] S343, mapping the identified color features and the clustering structure between the color features to the flower pattern feature data, and determining that the flower pattern sample corresponds to the color feature cluster by comparing the similarity between the flower pattern sample and the cluster center;

[0058] S344. Based on the mapping result, visually evaluate the identified color combinations and screen the color combinations.

[0059] Optionally, combining the color combination obtained from the color structure recognition model with the color combination analyzed and recognized based on the color psychology principle, and determining the best lining fabric pattern color combination based on the combination result comprises the following steps:

[0060] S361. Analyze the color combinations identified by the color structure recognition model, and compare and evaluate the color combinations with the content provided by the color psychology principles;

[0061] S362. Based on the results of the comparative evaluation of the model and psychology, integrate color data and obtain color preference information of target consumers collected through market research;

[0062] S363. Select a color combination that resonates with the target market based on the integrated color data and the color preference information of the target consumers, and conduct verification;

[0063] S364. Obtain the best verified interlining pattern and color combination and document it.

[0064] According to another aspect of the present invention, there is also provided a lining fabric pattern intelligent color separation system based on clustering and stratification, the system comprising: a feature extraction module, a feature clustering module and a color analysis and preference extraction module;

[0065] The feature extraction module is connected with the color analysis and preference extraction module through the feature clustering module;

[0066] A feature extraction module is used to obtain an interlining cloth image to be detected, pre-process the interlining cloth image to be detected, obtain an interlining cloth image with enhanced image quality, and extract interlining cloth feature data from the interlining cloth image with enhanced image quality;

[0067] A feature clustering module is used to cluster the lining feature data using a clustering algorithm and identify feature areas that meet a preset similarity threshold;

[0068] The color analysis and preference extraction module is used to extract the pattern feature data in the feature area, use the color structure recognition model to analyze the pattern feature data in the feature area and extract the color combination, and combine color psychology to extract the color combination preference to obtain the best lining cloth pattern color combination.

[0069] Compared with the prior art, this application has the following beneficial effects:

[0070] 1. The present invention can effectively identify and distinguish different feature areas in the lining cloth image, including color, texture, etc., through the spectral clustering algorithm, can effectively process complex or non-spherical data distribution, allow the selection of different similarity metrics, and provide a high degree of flexibility to adapt to different types of data features and analysis requirements. The eigendecomposition of the Laplace matrix is ​​used for data dimensionality reduction, which can more clearly reveal the intrinsic structure of the data and reduce the computational burden when processing complex data while retaining the most important data characteristics. The results of the spectral clustering algorithm can be mapped back to the original lining cloth image, so that each cluster is clearly marked as a different feature area, and visualization greatly enhances the interpretability and intuitiveness of the analysis results.

[0071] 2. The present invention can dynamically set the similarity threshold according to the specific characteristics of the data and the analysis requirements, allowing only sufficiently similar feature points to be clustered, thereby improving the accuracy and relevance of the clustering results. The flexibility of spectral clustering makes it applicable to various types of data, making it very useful in multiple fields. Through the precise clustering and mapping of the interlining feature areas, it can provide strong decision support for interlining design, production and quality control.

[0072] 3. The present invention can accurately identify and extract color combinations in specific patterns through the color structure recognition model, and extract primary and secondary colors from complex designs, providing an accurate data basis for subsequent analysis and application. Combined with the principles of color psychology, it can evaluate the visual and emotional effects of different color combinations, which is helpful to design products that meet the emotional needs of the target market and ensure that the selected color combination can resonate with consumers on a psychological level. By collecting and analyzing consumers' color preferences, the needs of the target market can be better understood, thereby selecting color combinations that can stimulate positive responses, thereby improving the market acceptance of products and consumer satisfaction.

[0073] 4. The present invention combines technical analysis with market research data, providing solid data support for the decision-making process, ensuring that design decisions are based on scientific analysis and close to market reality. Through the combination of models and psychological principles, designers can quickly iterate and optimize flower color designs, reduce trial and error costs, and accelerate product development cycles. Well-designed color combinations are not only visually attractive, but also emotionally connect with consumers, enhancing the market competitiveness of products. The final color combination is documented in detail, which helps to standardize the production process and ensure consistency in design quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The above characteristics, features and advantages of the present invention and their implementation methods and methods will become more clearly understood in conjunction with the following description of the embodiments, which are described in detail in conjunction with the accompanying drawings. Herein, a schematic diagram is shown:

[0075] Figure 1 is a flow chart of a method for intelligent color separation of lining fabric patterns based on clustering and stratification according to an embodiment of the present invention;

[0076] Figure 2 The present invention is a block diagram of a system for intelligent color separation of lining fabric patterns based on clustering and stratification according to an embodiment of the present invention.

[0077] In the figure:

[0078] 1. Feature extraction module; 2. Feature clustering module; 3. Color analysis and preference extraction module. DETAILED DESCRIPTION

[0079] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0080] According to an embodiment of the present invention, a method and system for intelligent color separation of lining fabric patterns based on clustering and stratification are provided.

[0081] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to one embodiment of the present invention, a method for intelligent color separation of lining fabric patterns based on clustering and stratification is provided, and the method for intelligent color separation of lining fabric patterns comprises the following steps:

[0082] S1. Acquire an interlining cloth image to be detected, pre-process the interlining cloth image to be detected, acquire an interlining cloth image with enhanced image quality, and extract interlining cloth feature data from the interlining cloth image with enhanced image quality.

[0083] It should be explained that a high-resolution camera or scanner is used to obtain a digital image of the lining cloth to ensure image clarity and uniform lighting for subsequent processing under a specific lighting control environment. Standardized shooting equipment is used to ensure consistency in image quality. The preliminary step of obtaining the image is the basis of the entire image processing process. High-quality original images can effectively improve the accuracy and reliability of subsequent processing. Digital image processing software or algorithms (such as median filtering, Gaussian filtering, etc.) are used to reduce random noise in the image, adjust the color balance and contrast of the image, and ensure color authenticity and consistency. If the image is not oriented correctly or contains unnecessary edge portions, rotation correction and cropping to the appropriate size and angle are required. Preprocessing is a key step to improve image quality and make it more suitable for feature extraction and analysis. The purpose of processing is to reduce error sources and enhance useful signals in the image.

[0084] Using techniques such as histogram equalization or contrast-limited adaptive histogram equalization (CLAHE) to enhance the local contrast and details of the image, image enhancement aims to improve the visual effect of the image and make the difficult-to-observe details more obvious. It uses edge detection (such as Sobel, Canny algorithm), texture analysis (such as gray-level co-occurrence matrix) or deep learning methods (such as convolutional neural networks) to extract key features in the image, such as texture, color distribution, shape, etc., and convert the extracted features into a data format suitable for machine learning model processing, such as feature vectors.

[0085] S2. Clustering the lining feature data using a clustering algorithm to identify feature areas that meet a preset similarity threshold.

[0086] Preferably, clustering the lining cloth feature data using a clustering algorithm to identify feature areas that meet a preset similarity threshold comprises the following steps:

[0087] S21. Acquire lining cloth feature data, and select a similarity measure of the lining cloth feature data.

[0088] Select an appropriate similarity measurement method based on the analysis requirements. Commonly used similarity metrics include Euclidean distance, cosine similarity, Manhattan distance, etc. The selected standard should match the type of feature and the analysis goal. Data analysis software or programming languages ​​(such as the SciPy library in Python) can be used to implement these metrics to evaluate and compare the feature similarity between different lining images.

[0089] S22. Calculate similarity based on the selected lining cloth feature data and similarity measurement, and construct a similarity matrix.

[0090] Preferably, calculating the similarity based on the selected interlining feature data and the similarity metric and constructing a similarity matrix comprises the following steps:

[0091] S221, selecting a similarity measure according to the extracted lining cloth feature data;

[0092] S222, for each pair of pixels in the lining cloth feature data, using a kernel function to calculate the similarity between feature vectors;

[0093] S223, pre-configuring a similarity threshold, and retaining similarities higher than the pre-configured similarity threshold;

[0094] S224, filling similarities higher than a pre-configured similarity threshold into the similarity matrix respectively, and constructing a similarity matrix;

[0095] Among them, the formula for calculating the similarity between feature vectors using the kernel function is:

[0096]

[0097] Where W ij Represents the similarity between pixel i and pixel j;

[0098] exp() represents the natural exponential function;

[0099] x i Represents the feature vector of pixel i;

[0100] x j represents the feature vector of pixel j;

[0101] σ represents the bandwidth parameter of the Gaussian kernel function;

[0102] ||x i -x j || 2 represents the feature vector x i and the eigenvector x j The square of the Euclidean distance between them.

[0103] It should be explained that the most appropriate similarity measurement method is selected according to the characteristics of the extracted lining feature data (such as the dimension and distribution characteristics of the data). For example, cosine similarity may be selected for color distribution, and Euclidean distance may be selected for shape features. The selected similarity measurement is implemented in a data analysis software or programming environment. For example, Python's SciPy or NumPy library is used to calculate different distance metrics.

[0104] Select a kernel function (such as Gaussian kernel, linear kernel, etc.). The kernel function can calculate the similarity between data points in high-dimensional space. Use the selected kernel function to calculate the similarity of each pair of feature vectors. This is usually implemented through loops or matrix operations in a programming environment. Set a reasonable similarity threshold based on the analysis purpose and data characteristics. The threshold is used to filter out sufficiently similar pairs of data points and only retain those results with a similarity higher than the set threshold. This helps reduce the amount of calculation and improve the focus and efficiency of subsequent analysis. Fill the filtered similarity data into a matrix. Usually, this matrix is ​​symmetrical, and each element of the matrix represents the similarity between a pair of data points.

[0105] Determine a suitable similarity threshold based on the characteristics of the data and the analysis requirements. The setting of the similarity threshold is based on prior statistical analysis or expert experience. The statistical analysis library in the programming environment (such as Python) is used to analyze the data distribution and determine a suitable threshold. All calculated similarity values ​​are traversed, and only those similarity data above the threshold are retained. This is usually achieved by writing conditional filtering statements. In programming languages ​​such as Python, array operations or loop statements can be used to implement filtering. For example, the conditional expression function of the NumPy library is used to fill the filtered similarity values ​​into a matrix according to the corresponding indexes. Usually, this matrix is ​​symmetrical, that is, the values ​​of the i, j positions and the j, i positions of the matrix are the same, both indicating the similarity between the i-th and j-th data points. Once the similarity matrix is ​​constructed, it can be saved as a file for subsequent data analysis tasks such as clustering, classification, or network analysis.

[0106] S23. Construct a Laplace matrix based on the similarity matrix, and perform eigendecomposition to obtain the eigenvalues ​​and eigenvectors of the Laplace matrix.

[0107] Preferably, constructing a Laplace matrix according to the similarity matrix and performing eigendecomposition to obtain the eigenvalues ​​and eigenvectors of the Laplace matrix includes the following steps:

[0108] S231, based on each row in the similarity matrix, calculate the sum of all elements in each row, and set the sum to the element on the corresponding diagonal of the degree matrix;

[0109] S232, constructing a Laplace matrix according to the similarity matrix and the degree matrix;

[0110] S233. Solve the eigenvalues ​​and eigenvectors of the Laplace matrix.

[0111] It should be explained that summing the elements of each row in a similar matrix is ​​easily accomplished through matrix operations in a programming language such as Python, using functions such as numpy.sum() to sum each row of the matrix and create a diagonal matrix whose diagonal elements are the sum of the elements in the corresponding rows. In Python, the diagonal matrix can be created directly from the sum array using the numpy.diag() function.

[0112] Each diagonal element of the degree matrix represents the connection strength of the corresponding node, that is, the sum of the similarities between the node and other nodes.

[0113] The Laplacian matrix L is usually defined as L=DW, where D is the degree matrix and W is the similarity matrix, and is implemented through simple matrix operations.

[0114] Use numerical algorithms to solve the eigenvalues ​​and eigenvectors of the Laplacian matrix. In Python, you can use numpy.linalg.eig() or scipy.linalg.eigh() to perform eigendecomposition of the Laplacian matrix.

[0115] Sort by the size of the eigenvalues, because the eigenvectors corresponding to small eigenvalues ​​can reveal the dense clustering structure of the data. You can use numpy's argsort() function to sort the eigenvalues ​​and rearrange the eigenvectors, and select an appropriate number of minimum non-zero eigenvalues ​​and their corresponding eigenvectors for the next step of analysis.

[0116] S24. According to the eigenvalues ​​and eigenvectors of the Laplace matrix, select the eigenvector corresponding to the minimum eigenvalue, and cluster the eigenvectors.

[0117] Preferably, according to the eigenvalues ​​and eigenvectors of the Laplace matrix, selecting the eigenvector corresponding to the minimum eigenvalue, and clustering the eigenvectors comprises the following steps:

[0118] S241, selecting an eigenvector with a minimum eigenvalue from the eigendecomposition of the Laplace matrix;

[0119] S242, performing standardization processing on the selected feature vector;

[0120] S243, clustering the standardized feature vectors;

[0121] S244, analyzing the clustering results, checking the quality of each cluster, and judging whether the division of feature regions is reasonable;

[0122] S245. Map each cluster back to the lining cloth image according to the judgment result, wherein each cluster corresponds to each feature area.

[0123] It should be explained that after completing the eigendecomposition of the Laplace matrix, the eigenvector with the smallest eigenvalue is selected, which usually means excluding the case where the eigenvalue is zero, and then the first eigenvector is selected. The selected eigenvector is standardized to ensure that each element is on the same scale. This is achieved by subtracting the mean value of the vector from each element and dividing it by the standard deviation. The standardized eigenvectors are clustered using a spectral clustering algorithm. The algorithm divides the eigenvectors into several groups, each representing a class of similar data points. The clustering results are analyzed, and indicators such as the silhouette coefficient are used to evaluate the quality and consistency of the clustering. In addition, the clustering results are visualized to check the separation and cohesion between the clusters, and the clustering results are mapped back to the original lining cloth image. Each cluster corresponds to a feature area in the image, and the cluster labels are matched to image pixels or areas.

[0124] Spectral clustering is a graph-based clustering method that uses the similarity matrix of data to perform clustering. It first reduces the dimension of the data through the eigenvectors of the Laplacian matrix, and then runs traditional clustering algorithms on these reduced eigenvectors.

[0125] Using Python, you can usually implement spectral clustering by importing numpy and scikit-learn libraries, determining the number of clusters you want, and selecting other parameters of the spectral clustering algorithm. The main parameters of spectral clustering include the number of clusters and the similarity measure. Apply the spectral clustering algorithm to the standardized feature vectors, which automatically performs the eigendecomposition of the Laplacian matrix, and then use the K-means algorithm to cluster in the resulting low-dimensional feature space. Analyze the clustering results, view the members and characteristics of each group, and use indicators such as the silhouette coefficient to evaluate the effectiveness of the clustering. Visualization tools can also help show the distribution of clustering results.

[0126] The example code for applying spectral clustering to a set of standardized feature vectors is as follows:

[0127] import numpy as np

[0128] from sklearn.cluster import SpectralClustering

[0129] #Assume that normalized_feature_vectors is the feature vector that has been standardized

[0130] normalized_feature_vectors = np.random.rand(100, 10) #generate some sample data

[0131] #Instantiate spectral clustering object

[0132] n_clusters = 3 #Suppose we want to divide the data into 3 clusters

[0133] clustering=SpectralClustering(n_clusters=n_clusters, affinity='nearest_neighbors')

[0134] # Apply spectral clustering

[0135] labels=clustering.fit_predict(normalized_feature_vectors)

[0136] # Output cluster labels

[0137] print("Cluster labels:",labels)

[0138] The above process reveals the intrinsic structure of the data through dimensionality reduction and distance-based clustering techniques, making spectral clustering particularly suitable for dealing with complex or non-spherical data distributions.

[0139] S25. According to the clustering result, the clustering result is mapped back to the lining cloth image, and each cluster is marked as a feature region.

[0140] S3. Extract the pattern feature data in the feature area, use the color structure recognition model to analyze the pattern feature data in the feature area and extract the color combination, and combine color psychology to extract the preference of color combination to obtain the best lining cloth pattern color combination.

[0141] Preferably, extracting the pattern feature data in the feature area, analyzing the pattern feature data in the feature area using a color structure recognition model and extracting color combinations, and extracting the color combination preference in combination with color psychology to obtain the best lining pattern color combination includes the following steps:

[0142] S31, acquiring pattern feature data from the feature area.

[0143] It should be explained that based on the clustering results of the previous step, the feature area containing the flower pattern design is determined. The feature area is represented by different clusters identified by the clustering algorithm. Each cluster corresponds to a specific area in the image. Image processing technology is used to extract flower pattern features from the determined feature area. Common flower pattern features include shape, color distribution, texture, etc. Edge detection, color analysis, texture descriptors and other methods can be used to extract these features; for example, the Sobel operator is used for edge detection, the color histogram is used to analyze the color distribution, and the local binary patterns (Local Binary Patterns, LBP) are used for texture analysis.

[0144] Convert the extracted features into a format suitable for machine learning model processing, such as feature vectors or feature matrices, standardize the feature data to eliminate the impact of different magnitudes and ranges, ensure the fairness and accuracy of the data during analysis, and store the encoded and standardized pattern feature data in a suitable format, such as a CSV file or database, to facilitate subsequent data analysis.

[0145] S32. Preconfigure the framework of the color structure recognition model and match the dimension of the flower pattern feature data.

[0146] It needs to be explained that the color structure recognition model is a self-organizing feature mapping network model. The self-organizing feature mapping network model is selected as the model for color structure recognition, the size and shape of the grid are determined (such as a 10x10 grid), the extracted flower pattern feature data is converted into a format suitable for the input of the self-organizing feature mapping network model, ensuring that all feature vectors have the same dimension and have been properly normalized, the self-organizing feature mapping network model is initialized according to the dimension of the feature data, and an appropriate learning rate and neighborhood function are set. These parameters will affect the learning process of the model and the final feature mapping. The flower pattern feature data is input into the self-organizing feature mapping network model for training. During the training process, the network will self-organize to form features mapped to a two-dimensional space.

[0147] S33. Input the flower pattern feature data into the color structure recognition model for training, and aggregate the feature parameters on adjacent neurons through self-organizing features.

[0148] Preferably, inputting the flower pattern feature data into the color structure recognition model for training, and aggregating the feature parameters on adjacent neurons through self-organizing features comprises the following steps:

[0149] S331, determining the size and shape of the color structure recognition model, and setting the initial weights of the neurons in the competition layer;

[0150] S332, normalizing the flower pattern feature data;

[0151] S333, for each flower pattern feature data, calculating the Euclidean distance between the flower pattern feature data and the neurons in the competition layer, and determining the winning neuron;

[0152] S334, updating the weights of the winning neuron and the neurons in the neighborhood of the winning neuron according to the learning rate and the neighborhood function;

[0153] S335, repeating the steps of S333-S334, performing multiple iterations on all the flower pattern feature data, until the color structure recognition model completes a predetermined number of iterations.

[0154] It should be explained that the weight of each neuron is randomly initialized, and the dimension of the weight is the same as the feature dimension of the input data. The input data is normalized to ensure that each feature dimension has a similar scale, which helps to improve training efficiency and model performance. For each input vector, its Euclidean distance with all neurons is calculated, and the neuron with the smallest distance is selected as the winning neuron. The weights of the winning neuron and the neurons in its neighborhood are updated according to the learning rate and neighborhood radius. The adjustment of weights depends on the difference between the input vector and the current weight. As the training progresses, the neighborhood radius and learning rate are gradually reduced to refine the learning and adaptation of the network. The entire data set is traversed multiple times, and the weights of the neurons are adjusted each iteration until the preset number of iterations is reached, or the network stabilizes and the changes are very small.

[0155] For each flower pattern feature data, the Euclidean distance between the flower pattern feature data and the neurons in the competition layer is calculated, and the specific steps to determine the winning neuron are as follows:

[0156] Prepare input data and neuron weight matrices,

[0157] The input data is normalized pattern feature data, which is a feature vector.

[0158] The weight matrix is ​​the neuron weight in the self-organizing feature map network model. The weight of each neuron is also a vector with the same dimension as feature_vector.

[0159] For each flower pattern feature vector input, the Euclidean distance between the flower pattern feature vector and each neuron weight vector in the self-organizing feature map network model is calculated. The calculation formula is:

[0160]

[0161] Where n represents the dimension of the vector;

[0162] a represents the input flower pattern feature vector;

[0163] b represents the weight vector of a neuron in the self-organizing feature map network model;

[0164] a i Represents the component of the input flower pattern feature vector a in the i-th dimension;

[0165] n i Represents the component of the weight vector b of a neuron in the self-organizing feature map network model in the i-th dimension;

[0166] d(a, b) represents the Euclidean distance between the flower pattern feature vector a and the weight vector b of a neuron in the self-organizing feature map network model.

[0167] The winning neuron (Best Matching Unit, BMU) is the neuron with the smallest distance to the current flower pattern feature vector. In simple terms, it is to find out which neuron's weight is closest to the input feature vector among all neurons;

[0168] The example code for how to calculate the winning neuron is as follows:

[0169]

[0170]

[0171] Through step S333, the self-organizing feature mapping network model can find the neuron (BMU) that is most similar to the current input feature.

[0172] According to the learning rate and neighborhood function, the specific steps to update the weights of the winning neuron and the neurons in the neighborhood of the winning neuron are as follows:

[0173] Determine the best matching unit (BMU) of the input vector, that is, the neuron closest to the input vector. Calculate the neighborhood radius according to the progress of training. Generally, the neighborhood radius decreases over time to refine the learning process. Calculate the current learning rate, which generally decreases with the number of iterations. For each neuron in the BMU and its neighborhood, update its weight to be closer to the input vector. Through these steps, the SOM network gradually adjusts the weights of the neurons so that the network can capture and represent the structural characteristics of the input data.

[0174] S34. Analyze the output of the color structure recognition model to identify color combinations.

[0175] Preferably, analyzing the output of the color structure recognition model to identify the color combination comprises the following steps:

[0176] S341, after the training is completed, extracting the weight vector of each neuron from the competition layer of the color structure recognition model as output data;

[0177] S342, analyzing the extracted output data to identify different color features and clustering structures between the color features;

[0178] S343, mapping the identified color features and the clustering structure between the color features to the flower pattern feature data, and determining that the flower pattern sample corresponds to the color feature cluster by comparing the similarity between the flower pattern sample and the cluster center;

[0179] S344. Based on the mapping result, visually evaluate the identified color combinations and screen the color combinations.

[0180] It needs to be explained that after the training is completed, the weight vector of each neuron is extracted from the competitive layer of the color structure recognition model. The weight vector represents the clustering center of different color features learned by the network. The extracted weight vector is analyzed to identify different color features. The clustering algorithm is used to deeply analyze the similarities between these weight vectors to form a clustering structure of color features. The feature data of the flower pattern sample is compared with the extracted clustering center to determine which color feature clusters the flower pattern sample matches. This usually involves calculating the distance between the sample and each cluster center to find the most similar cluster. Based on the mapping results, the identified color combinations are visually evaluated to screen out the most representative and visually attractive color combinations.

[0181] The clustering structure is a logical grouping of color features formed through network self-organization. The clustering structure reflects the inherent patterns and similarities in the input data (such as flower pattern feature data).

[0182] The identified color features and the clustering structure between the color features are mapped to the flower pattern feature data. By comparing the similarity between the flower pattern sample and the cluster center, the specific steps to determine the correspondence between the flower pattern sample and the color feature cluster are as follows:

[0183] After the training process of the color structure recognition model is completed, the weight vector of each neuron represents a cluster center, which captures the typical color features in the training data. For each flower pattern sample feature vector, its distance from all cluster centers is calculated. Euclidean distance or other similarity measurement methods can be used to select the cluster center with the smallest distance for each flower pattern sample as the most matching cluster. The matching process indicates which color feature cluster the sample is closest to. Each flower pattern sample is mapped to the corresponding cluster based on its matching result with the cluster center. The mapping result can be used for further analysis and visualization to help understand the color relationship and distribution between samples. Mapping not only helps to identify and understand the main color patterns in flower pattern data, but also provides a basis for further design innovation and product development.

[0184] S35. Analyze and identify color combinations based on the principles of color psychology.

[0185] It should be explained that color psychology studies how colors affect people's emotions and behaviors. The purpose is to use the knowledge of color psychology to enhance the emotional and psychological effects of pattern design. The principles of color psychology point out that different colors can cause different emotional responses in people; for example, blue is often seen as a calm and reassuring color, while red may inspire energy and passion. Applying these principles, pattern design can be evaluated and optimized to produce the desired emotional effect in the target market or consumers.

[0186] From the output of the color structure recognition model, specific color combinations have been identified. The color combinations are obtained through a clustering process and reflect the common color patterns in the data. The psychological impact of each color is studied. This includes the impact of color on emotions, the significance of color and cultural associations, and the role of color in specific environments. For example, the calming effect of green in natural products, or the lively, happy effect of yellow in children's products, is studied. Each identified color combination is analyzed to predict the psychological and emotional effects they may have in visual communication. The utility of color combinations in specific application scenarios (such as clothing, interior decoration, advertising, etc.) is considered. It is determined whether the color combination meets the expected emotional response of the target market. For example, it is determined whether a set of color combinations dominated by soft blue and green is suitable for advertising for healthcare brands. Based on the analysis of color psychology, the pattern design is adjusted and optimized to maximize its emotional impact, including adjusting color intensity, contrast or color scheme. Based on feedback and further research on the principles of color psychology, the effectiveness of the design is evaluated and iterative improvements are made as needed.

[0187] S36, combining the color combination obtained from the color structure recognition model with the color combination analyzed and recognized based on the color psychology principle, and determining the best lining fabric pattern color combination based on the combination result.

[0188] Preferably, the color combination obtained from the color structure recognition model is combined with the color combination analyzed and identified based on the color psychology principle, and based on the combination result, determining the best lining fabric pattern color combination comprises the following steps:

[0189] S361. Analyze the color combinations identified by the color structure recognition model, and compare and evaluate the color combinations with the content provided by the color psychology principles;

[0190] S362. Based on the results of the comparative evaluation of the model and psychology, integrate color data and obtain color preference information of target consumers collected through market research;

[0191] S363. Select a color combination that resonates with the target market based on the integrated color data and the color preference information of the target consumers, and conduct verification;

[0192] S364. Obtain the best verified interlining pattern and color combination and document it.

[0193] It should be explained that we first analyze the color combinations identified by the color structure recognition model. The color combinations represent the main color patterns in the data. We then compare the color combinations with the principles of color psychology, evaluate the emotional responses that each combination may cause, and merge the color analysis results from the color structure recognition model with the psychological evaluation results to form a comprehensive set of color data. We then conduct market research and collect color preference information from target consumers. This can be achieved through questionnaires, social media analysis, etc. We then select the color combination that is most likely to resonate with the target market based on the integrated color data and consumer preferences. We then verify the market acceptance of the selected color combination through methods such as focus groups, A / B testing, or prototype product testing. We then document the final color combination in detail, including design concepts, color application guidelines, and market test results.

[0194] In addition, the weight vectors of each neuron are extracted from the color structure recognition model. The weight vectors represent the color combinations learned by the model during the training process. The weight vectors are converted into visual color patterns or directly interpreted as color combinations for analysis, research and collection of psychological principles on how different colors affect emotions and behaviors, including how colors affect people's moods, their ability to stimulate specific emotions, and their cultural significance.

[0195] Documentation refers to the process of recording and organizing processes, results and important information for future reference, review or compliance with standards. Documentation in the field of color combination and design usually includes the following key aspects:

[0196] 1) Design specifications and parameter records

[0197] Record the specific specifications of the design, such as the RGB values ​​of the color combination, the type of materials used, the design dimensions and layout parameters, etc.

[0198] Detailed description of the design software and version used to ensure compatibility when revisiting or modifying.

[0199] 2) Design decisions and rationale

[0200] Document all factors considered in the design decision process, including market research, consumer preferences, and psychological influences.

[0201] Describe in detail why specific color combinations and design elements were chosen, including the data analysis and psychology behind them.

[0202] 3) Results and performance evaluation

[0203] Record the final effects of the design, such as visual presentation, consumer feedback, and market performance.

[0204] This includes any form of test results or performance evaluation reports, which may be based on simulation demonstrations or feedback from actual applications.

[0205] 4) Standard operating procedures and maintenance guidelines

[0206] Provide necessary operating manuals and guidelines to ensure standardization of designs during production and implementation.

[0207] Document any special requirements or considerations so they can be followed during production and later maintenance of the design.

[0208] 5) History of Improvement and Iteration

[0209] Record the iterative process of the design from the first draft to the final version, including the intermediate modification, adjustment and optimization steps.

[0210] The version number and modification date should be clearly marked, and the purpose and results of each iteration should also be recorded in detail.

[0211] According to another embodiment of the present invention, Figure 2 As shown, a lining fabric pattern intelligent color separation system based on clustering and stratification is also provided, the system comprising: a feature extraction module 1, a feature clustering module 2 and a color analysis and preference extraction module 3;

[0212] The feature extraction module 1 is connected to the color analysis and preference extraction module 3 through the feature clustering module 2;

[0213] The feature extraction module 1 is used to obtain the interlining cloth image to be detected, pre-process the interlining cloth image to be detected, obtain the interlining cloth image with enhanced image quality, and extract interlining cloth feature data from the interlining cloth image with enhanced image quality;

[0214] Feature clustering module 2, used to cluster the lining feature data using a clustering algorithm and identify feature areas that meet a preset similarity threshold;

[0215] The color analysis and preference extraction module 3 is used to extract the pattern feature data in the feature area, use the color structure recognition model to analyze the pattern feature data in the feature area and extract the color combination, and at the same time combine color psychology to extract the preference of color combination to obtain the best lining cloth pattern color combination.

[0216] In summary, with the help of the above technical solution of the present invention, the present invention can effectively identify and distinguish different feature areas in the lining cloth image, including color, texture, etc., through the spectral clustering algorithm, can effectively process complex or non-spherical data distribution, allow the selection of different similarity metrics, and provide a high degree of flexibility to adapt to different types of data features and analysis requirements. The eigendecomposition of the Laplace matrix is ​​used for data dimensionality reduction, which can more clearly reveal the inherent structure of the data, reduce the computational burden when processing complex data, and retain the most important data characteristics. The result of the spectral clustering algorithm can be mapped back to the original lining cloth image, so that each cluster is clearly marked as a different feature area, and visualization greatly enhances the interpretability and intuitiveness of the analysis results. The present invention can dynamically set the similarity threshold according to the specific characteristics of the data and the analysis requirements, allowing only sufficiently similar feature points to be clustered, thereby improving the accuracy and relevance of the clustering results. The flexibility of spectral clustering makes it applicable to various types of data, making it very useful in multiple fields, and through the precise clustering and mapping of the lining cloth feature area, it can provide strong decision support for lining cloth design, production and quality control; the present invention uses the color structure recognition model to identify the similarity threshold of the data. The model can accurately identify and extract color combinations in specific flower patterns, and extract primary and secondary colors from complex designs, providing an accurate data basis for subsequent analysis and application. Combined with the principles of color psychology, the visual and emotional effects of different color combinations can be evaluated, which helps to design products that meet the emotional needs of the target market and ensure that the selected color combinations can resonate with consumers at the psychological level. By collecting and analyzing consumers' color preferences, the needs of the target market can be better understood, so as to select color combinations that can stimulate positive reactions, thereby improving the market acceptance and consumer satisfaction of the product. The present invention combines technical analysis with market research data to provide solid data support for the decision-making process, ensuring that design decisions are based on scientific analysis and close to market reality. Through the combination of models and psychological principles, designers can quickly iterate and optimize flower pattern color design, reduce trial and error costs, and accelerate product development cycles. Well-designed color combinations are not only visually attractive, but also emotionally connect with consumers, enhancing the market competitiveness of products. The final color combination is documented in detail, which helps to standardize the production process and ensure the consistency of design quality.

[0217] Although the present invention has been disclosed as above with preferred embodiments, the embodiments are merely examples for the convenience of description and are not intended to limit the present invention. Those skilled in the art may make several changes and modifications without departing from the spirit and scope of the present invention. The scope of protection claimed by the present invention shall be based on the claims.

Claims

1. A method for intelligent color separation of lining fabric patterns based on clustering and stratification, characterized in that: The lining fabric pattern intelligent color separation method comprises the following steps: S1, obtaining an interlining cloth image to be detected, preprocessing the interlining cloth image to be detected, obtaining an interlining cloth image with enhanced image quality, and extracting interlining cloth feature data from the interlining cloth image with enhanced image quality; S2, clustering the lining feature data using a clustering algorithm to identify feature areas that meet a preset similarity threshold; S3, extracting the flower pattern feature data in the feature area, analyzing the flower pattern feature data in the feature area using a color structure recognition model and extracting color combinations, and extracting the color combination preference in combination with color psychology to obtain the best lining cloth flower pattern color combination; The step of extracting the pattern feature data in the feature area, analyzing the pattern feature data in the feature area using a color structure recognition model and extracting color combinations, and extracting the color combination preference in combination with color psychology to obtain the best lining pattern color combination comprises the following steps: S31, acquiring flower pattern feature data from the feature area; S32, pre-configuring the framework of the color structure recognition model and matching the dimension of the flower pattern feature data; S33, inputting the flower pattern feature data into the color structure recognition model for training, and clustering the feature parameters on adjacent neurons through self-organizing features; S34, analyzing the output of the color structure recognition model to identify the color combination; S35. Analyze and identify color combinations based on the principles of color psychology; S36, combining the color combination obtained from the color structure recognition model with the color combination analyzed and recognized based on the color psychology principle, and determining the best lining fabric pattern color combination based on the combination result; The step of inputting the flower pattern feature data into the color structure recognition model for training, and aggregating the feature parameters on adjacent neurons through self-organizing features comprises the following steps: S331, determining the size and shape of the color structure recognition model, and setting the initial weights of the neurons in the competition layer; S332, normalizing the flower pattern feature data; S333, for each flower pattern feature data, calculating the Euclidean distance between the flower pattern feature data and the neurons in the competition layer, and determining the winning neuron; S334, updating the weights of the winning neuron and the neurons in the neighborhood of the winning neuron according to the learning rate and the neighborhood function; S335, repeating the steps of S333-S334, performing multiple iterations on all flower pattern feature data until the color structure recognition model completes a predetermined number of iterations; Analyzing the output of the color structure recognition model and identifying the color combination comprises the following steps: S341, after the training is completed, extracting the weight vector of each neuron from the competition layer of the color structure recognition model as output data; S342, analyzing the extracted output data to identify different color features and clustering structures between the color features; S343, mapping the identified color features and the clustering structure between the color features to the flower pattern feature data, and determining that the flower pattern sample corresponds to the color feature cluster by comparing the similarity between the flower pattern sample and the cluster center; S344, based on the mapping result, visually evaluate the identified color combination and screen the color combination; The method of combining the color combination obtained from the color structure recognition model with the color combination analyzed and recognized based on the color psychology principle, and determining the best lining fabric pattern color combination based on the combination result comprises the following steps: S361. Analyze the color combinations identified by the color structure recognition model, and compare and evaluate the color combinations with the content provided by the color psychology principles; S362. Based on the results of the comparative evaluation of the model and psychology, integrate color data and obtain color preference information of target consumers collected through market research; S363. Select a color combination that resonates with the target market based on the integrated color data and the color preference information of the target consumers, and conduct verification; S364. Obtain the best verified interlining pattern and color combination and document it.

2. The method for intelligent color separation of lining fabric patterns based on clustering and stratification according to claim 1, characterized in that: The method of clustering the lining cloth feature data using a clustering algorithm and identifying feature areas that meet a preset similarity threshold comprises the following steps: S21, obtaining lining cloth feature data, and selecting a similarity measure of the lining cloth feature data; S22, calculating similarity based on the selected lining cloth feature data and similarity measurement, and constructing a similarity matrix; S23, constructing a Laplace matrix according to the similarity matrix, and performing eigendecomposition to obtain eigenvalues ​​and eigenvectors of the Laplace matrix; S24, according to the eigenvalues ​​and eigenvectors of the Laplace matrix, selecting the eigenvector corresponding to the minimum eigenvalue, and clustering the eigenvectors; S25. According to the clustering result, the clustering result is mapped back to the lining cloth image, and each cluster is marked as a feature region.

3. The method for intelligent color separation of lining fabric patterns based on clustering and stratification according to claim 2, characterized in that: The method of calculating similarity based on the selected interlining feature data and similarity measurement and constructing a similarity matrix comprises the following steps: S221, selecting a similarity measure according to the extracted lining cloth feature data; S222, for each pair of pixels in the lining cloth feature data, using a kernel function to calculate the similarity between feature vectors; S223, pre-configuring a similarity threshold, and retaining similarities higher than the pre-configured similarity threshold; S224, filling similarities higher than a pre-configured similarity threshold into the similarity matrix respectively, and constructing a similarity matrix; The formula for calculating the similarity between feature vectors using the kernel function is: ; In the formula, Represents pixels i and pixels j The similarity between represents the natural exponential function; Represents pixels i The eigenvector of Represents pixels j The eigenvector of represents the bandwidth parameter of the Gaussian kernel function; Represents the feature vector x i and the eigenvector x j The square of the Euclidean distance between them.

4. The method for intelligent color separation of lining fabric patterns based on clustering and stratification according to claim 3, characterized in that: The step of constructing a Laplace matrix according to a similarity matrix and performing eigendecomposition to obtain eigenvalues ​​and eigenvectors of the Laplace matrix includes the following steps: S231, based on each row in the similarity matrix, calculate the sum of all elements in each row, and set the sum to the element on the corresponding diagonal of the degree matrix; S232, constructing a Laplace matrix according to the similarity matrix and the degree matrix; S233. Solve the eigenvalues ​​and eigenvectors of the Laplace matrix.

5. The method for intelligent color separation of lining fabric patterns based on clustering and stratification according to claim 4, characterized in that: The step of selecting the eigenvector corresponding to the minimum eigenvalue according to the eigenvalues ​​and eigenvectors of the Laplace matrix and clustering the eigenvectors comprises the following steps: S241, selecting an eigenvector with a minimum eigenvalue from the eigendecomposition of the Laplace matrix; S242, performing standardization processing on the selected feature vector; S243, clustering the standardized feature vectors; S244, analyzing the clustering results, checking the quality of each cluster, and judging whether the division of feature regions is reasonable; S245. Map each cluster back to the lining cloth image according to the judgment result, wherein each cluster corresponds to each feature area.

6. A system for intelligent color separation of lining fabric patterns based on clustering and stratification, used to implement the method for intelligent color separation of lining fabric patterns based on clustering and stratification as claimed in any one of claims 1 to 5, characterized in that: The system includes: a feature extraction module, a feature clustering module and a color analysis and preference extraction module; The feature extraction module is connected with the color analysis and preference extraction module through the feature clustering module; The feature extraction module is used to obtain the interlining cloth image to be detected, pre-process the interlining cloth image to be detected, obtain the interlining cloth image with enhanced image quality, and extract interlining cloth feature data from the interlining cloth image with enhanced image quality; The feature clustering module is used to cluster the lining feature data using a clustering algorithm to identify feature areas that meet a preset similarity threshold; The color analysis and preference extraction module is used to extract the flower pattern feature data in the feature area, use the color structure recognition model to analyze the flower pattern feature data in the feature area and extract color combinations, and extract the color combination preferences in combination with color psychology to obtain the best lining cloth flower pattern color combination.