A method for optimizing clothing pattern making parameters based on big data
Through multi-channel data collection and deep learning algorithms, a comprehensive data model is built, which solves the problems of complex data integration and low optimization efficiency in traditional clothing plate making, realizes the automation and intelligence of clothing plate making parameters, and improves the integration and comfort of clothing. Spend.
Patent Information
- Application Number
- CN202411244772.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-09-06
AI Technical Summary
The traditional clothing plate making process relies on designer experience and a single data source, which is difficult to meet personalized needs, and the data integration and processing are complex, resulting in low efficiency and low accuracy in plate making parameter optimization, and it is impossible to effectively respond to the multi-dimensional parameter optimization needs.
A method based on big data is adopted to build a comprehensive data model through multi-channel data collection and deep fusion, and nonlinear relationships are extracted using generative adversarial networks and deep learning algorithms, and dynamic simulation verification and optimization are carried out to realize automatic adjustment and optimization of plate making parameters.
It significantly improves the optimization efficiency and accuracy of clothing plate making parameters, meets the personalized needs of different users, improves the fit and comfort of clothing, and realizes the automation and intelligence of the plate making process.
Smart Images

Figure CN118761174B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of clothing, and in particular to a clothing pattern making parameter optimization method based on big data. Background Art
[0002] With the rapid development of the clothing industry, consumers have put forward higher and higher requirements for the fit and comfort of clothing. However, the traditional clothing pattern making process mainly relies on the designer's experience and manual adjustment, which is not only inefficient but also difficult to accurately meet the personalized needs of different users. The existing technology has the following main defects in the optimization of clothing pattern making parameters:
[0003] In the existing technology, data mainly comes from a single channel, such as the experience data of fashion designers or a small amount of market research data. These data are small and incomplete, and it is difficult to reflect the real needs of different users and market trends. In addition, the data format is not unified, which makes data integration and processing complicated, affecting the effect of plate making parameter optimization. Multi-channel data collection and standardized processing is still a major challenge, and the comprehensiveness and consistency of data are difficult to guarantee.
[0004] Traditional clothing pattern making mainly relies on simple statistical analysis and the designer's experience and judgment, and cannot fully utilize big data analysis technology to explore the deep patterns and features hidden in the data. This leads to a lack of scientific basis for the adjustment process of pattern making parameters, which is prone to deviations and cannot meet personalized needs. In the existing technology, the limitations of data analysis tools and methods make it impossible to extract key factors from the complex data, and the optimization accuracy of pattern making parameters is limited.
[0005] In the prior art, the optimization of plate-making parameters usually adopts a simple trial-and-error method or a small number of linear regression models. These methods have limited effect when dealing with complex nonlinear relationships, and it is difficult to find the best combination of plate-making parameters. In addition, traditional algorithms require repeated experiments during the optimization process, which is time-consuming, labor-intensive, and inefficient. Complex body shape, behavior, and environmental data require more advanced algorithms to process, but the algorithm selection and application in the prior art are relatively single, and cannot effectively respond to multi-dimensional parameter optimization needs.
[0006] The existing pattern-making process usually lacks effective simulation verification methods, and the adjustment of pattern-making parameters mainly relies on the designer's experience and a small amount of actual fitting feedback. This not only increases the cost of trial and error, but may also cause the pattern-making parameters to fail to achieve the expected effect in actual production, affecting the fit and comfort of the garment. In the existing technology, the inadequacies of simulation models and verification methods limit the reliability and feasibility of parameter optimization solutions, resulting in poor application results in actual production.
[0007] In order to overcome these defects in the existing technology, it is necessary to propose a clothing pattern making parameter optimization method based on big data and intelligent optimization algorithm to improve the efficiency and accuracy of the pattern making process. Summary of the invention
[0008] One purpose of the present invention is to propose a method for optimizing clothing pattern making parameters based on big data. The present invention overcomes the shortcomings of the prior art in data acquisition, analysis, optimization and verification through systematic and intelligent means, and provides an efficient and accurate solution for clothing pattern making parameter optimization.
[0009] A method for optimizing clothing pattern making parameters based on big data according to an embodiment of the present invention comprises the following steps:
[0010] S1. Obtaining raw data related to clothing pattern making from multiple specific channels;
[0011] S2, using a data fusion algorithm to deeply fuse clothing pattern making parameter sets from different sources to obtain a clothing pattern making parameter set;
[0012] S3, constructing a comprehensive data model through the clothing pattern making parameter set, the comprehensive data model includes standardized body shape data, individual behavior data and environmental factors;
[0013] S4, enhancing the comprehensive data model and generating virtual samples through a generative adversarial network;
[0014] S5. Apply deep learning algorithms to extract patterns and features from the comprehensive data model, identify nonlinear relationships that affect platemaking parameters, and obtain prediction results of the comprehensive data model;
[0015] S6. Based on the prediction results, a dynamic simulation environment is constructed to simulate the effects of different pattern making parameters on the fit and comfort of clothing, allowing users to try on clothing in a virtual environment;
[0016] S7, adjust the plate making parameters according to the feedback from dynamic simulation and virtual fitting, apply the optimized parameters to actual production, and perform secondary optimization by continuously monitoring the feedback data;
[0017] S8. Collect feedback data from the actual production process, including users’ actual wearing experience and quality inspection results of the production line, and iteratively improve the optimization parameters.
[0018] Optionally, the S1 includes the following steps:
[0019] S11. Obtain users’ evaluations on the fit, comfort and style of clothing, and construct a user market feedback dataset M:
[0020] ;
[0021] in, is the i-th market feedback data sample, which includes users’ evaluations on the fit, comfort, and style of clothing;
[0022] S12. Obtain the chest circumference, waist circumference and hip circumference measured by the user, and construct the user body shape dataset U:
[0023] ;
[0024] in, is the body data sample of the jth user, which includes chest circumference, waist circumference and hip circumference;
[0025] S13. Obtain fabric types, process parameters, and quality inspection results for garment production, and construct a garment production dataset:
[0026] ;
[0027] in, is the kth production data sample, which includes fabric type, process parameters and quality inspection results.
[0028] Optionally, S2 includes the following steps:
[0029] S21, performing data cleaning, deduplication and standardization on the market feedback dataset M, the user body shape dataset U and the production dataset P;
[0030] S22, use the linear discriminant analysis algorithm to deeply fuse the standardized market feedback dataset M', the standardized user body shape dataset U' and the standardized production dataset P'. The linear discriminant analysis algorithm calculates the inter-class scatter matrix between each dataset. and the intra-class scatter matrix , determine the best projection direction and obtain the fused clothing pattern making parameter set F:
[0031] ;
[0032] Among them, LDA represents the linear discriminant analysis process, and outputs the fused clothing pattern making parameter set F.
[0033] Optionally, S3 includes the following steps:
[0034] S31, preliminarily processing the fused clothing pattern making parameter set F to extract standardized body shape data, individual behavior data and environmental factors;
[0035] ;
[0036] Each It contains a combination of physical data, behavioral data, and environmental factors;
[0037] S32, constructing the extracted data into a comprehensive data model G;
[0038] S33. Use the improved multimodal clustering algorithm to perform cluster analysis on the standardized body shape data T, identify different body shape groups by combining the geometric characteristics and mechanical properties of the body shape data, and construct a body shape clustering model K:
[0039] ;
[0040] in, Cluster centers representing normalized body size patterns;
[0041] By minimizing the geometric feature error and mechanical property errors The weighted sum of is used for clustering optimization:
[0042] ;
[0043] Among them, the geometric feature error :
[0044] ;
[0045] Mechanical property error :
[0046] ;
[0047] in, and is the weight parameter, represents the i-th body type data sample, represents the mechanical property data of the i-th body type sample, and They represent the geometric characteristics and mechanical properties of the kth cluster center respectively;
[0048] S34, using a behavior feature extraction algorithm to perform pattern recognition on individual behavior data B, combining the time and frequency characteristics of the behavior data, maximizing feature entropy to perform pattern recognition on user behavior feature patterns, and constructing a behavior feature model H:
[0049] ;
[0050] The process of constructing the behavioral feature model is to maximize the feature entropy To optimize:
[0051] ;
[0052] in, Indicates behavioral characteristics The probability distribution of
[0053] S35, using a multimodal fusion algorithm based on graph neural network to fuse the body clustering model K, behavioral feature model H and environmental factor E to construct a comprehensive data model G, and minimize the graph structure loss function Train the comprehensive data model G:
[0054] ;
[0055] Among them, the graph structure loss function Defined by combining node features and edge weights:
[0056] ;
[0057] Where |E| is the number of edges, and Represent the feature representation of nodes u and v respectively, is the global feature mean, is the regularization parameter;
[0058] A comprehensive data model G reflecting the impact of different body shapes, behavioral habits and environmental factors on clothing pattern making parameters is constructed through a multimodal fusion algorithm based on graph neural network.
[0059] Optionally, the comprehensive data model includes the following components:
[0060] Standardized body shape data T, including the user's chest circumference, waist circumference, hip circumference, shoulder width and height;
[0061] Individual behavior data B, including the user's daily activity level, exercise habits, and clothing preferences;
[0062] Environmental factors E include season, temperature and humidity.
[0063] Optionally, S4 includes the following steps:
[0064] S41, preprocessing the comprehensive data model G, dividing the data of the comprehensive data model G into a training set and validation set ;
[0065] S42. Construct a multi-level generative adversarial network, which includes a multi-level generator network for generating virtual samples. and a multi-layer discriminator network for distinguishing real samples from virtual samples ;
[0066] S43. Use training set Train a multi-level generative adversarial network to enable the generator to generate realistic virtual samples by minimizing the adaptive loss function of the generator and the multi-level discriminator network. Defined as:
[0067] ;
[0068] in, is the loss function of the generator, represents the expectation of a random noise vector, is the loss weight parameter, represents the feature loss function, which is optimized by calculating the feature difference between virtual samples and real samples;
[0069] Adaptive loss function of the discriminator Defined as:
[0070] ;
[0071] in, is the loss function of the discriminator, represents the expectation of real data, is the loss weight parameter, represents the feature loss function, which is optimized by calculating the feature difference between real samples and virtual samples;
[0072] S44. Through iterative training, use the training set Update the parameters of the generator and discriminator and use the validation set Evaluate model performance until the generator generates virtual samples With real samples Inability to distinguish;
[0073] S45. Construct a feedback loop generative adversarial network based on a multi-level generative adversarial network, and introduce a feedback mechanism from users and experts in the process of generating virtual samples;
[0074] Users and experts evaluate the generated virtual samples in real time through the interface and provide feedback. The feedback loop generative adversarial network dynamically adjusts the parameters of the generator and discriminator based on the feedback.
[0075] User and expert feedback As an adjustment factor, add it to the loss function of the generator and discriminator:
[0076] ;
[0077] ;
[0078] in, represents the expectation of real data, is the feedback adjustment coefficient, which indicates the weight of the feedback on the loss function. represents the expected value of user and expert feedback, represents the expectation of a random noise vector, is the feedback adjustment coefficient, which indicates the weight of the feedback on the loss function. represents the expected value of user and expert feedback, Express expectations of user and expert feedback, user and expert feedback It is expressed as:
[0079] ;
[0080] in, represents the i-th feedback, and N represents the total number of feedbacks;
[0081] S46, the generated virtual sample Add it to the comprehensive data model G, expand the data set G' of the comprehensive data model, and the expanded comprehensive data model G' is expressed as:
[0082] ;
[0083] Among them, G' represents the expanded comprehensive data model, which includes the original comprehensive data model G and the generated virtual samples ;
[0084] S47, post-processing the expanded comprehensive data model G' to remove outliers and noise, and obtain the final comprehensive data model G'' after enhancement processing.
[0085] Optionally, the multi-level generator network It consists of multiple levels of fully connected layers, convolutional layers, and attention mechanisms. The input is a random noise vector z and a comprehensive data model G. Virtual samples are generated through a multi-level generation structure. :
[0086] ;
[0087] Among them, z represents a random noise vector, which is the initial input of the generator. represents the generated virtual sample;
[0088] The multi-level generation structure includes a basic generation layer, a high-level generation layer, and a detail generation layer. Each layer is enhanced by skip connections and attention mechanisms. The specific structure is as follows:
[0089] ;
[0090] Among them, x is the input feature map, represents the convolution operation, represents the activation function, Represents the attention mechanism, which is used to enhance feature representation;
[0091] Multi-layer discriminator network It consists of multiple levels of convolutional layers, fully connected layers, and adaptive feature extraction layers, and the input is real samples. and virtual samples , output the distinction result D(G):
[0092] .
[0093] Optionally, S5 includes the following steps:
[0094] S51. Construct a multimodal deep learning network model, including a multi-layer convolutional neural network and a long short-term memory network, to extract patterns and features from the comprehensive data model and identify nonlinear relationships that affect platemaking parameters;
[0095] S53, multi-layer convolutional neural network extracts the spatial features of the comprehensive data model. The structure includes input layer, multiple convolutional layers, pooling layer and fully connected layer:
[0096] The input layer receives the combined input of standardized body shape data T, individual behavior data B and environmental factors E into the comprehensive data model G''. The convolution layer extracts spatial features through convolution operations. The pooling layer performs feature dimensionality reduction through maximum pooling or average pooling. The fully connected layer maps the output features of the convolution layer and the pooling layer to a high-dimensional space, and the output spatial feature representation ;
[0097] S54, long short-term memory network extracts the time series characteristics of the comprehensive data model, and the structure includes input layer, long short-term memory unit and output layer:
[0098] The input layer receives spatial feature representation As input, the long short-term memory unit extracts time series features through the operation of forget gate, input gate and output gate, and outputs time feature representation ;
[0099] S55. Combine multi-layer convolutional neural network and long short-term memory network, perform feature fusion through hybrid attention mechanism, and generate the final feature representation :
[0100] ;
[0101] in, and is the weight parameter, which is dynamically adjusted through training;
[0102] S56, using the improved non-negative matrix factorization method, combining feature importance and feature sparsity to represent the final feature To reduce the dimension, the goal is to find non-negative matrices W and H such that:
[0103] ;
[0104] Among them, W is the feature weight matrix, H is the feature matrix;
[0105] S57. The selected features are input into the plate-making parameter optimization model. The plate-making parameter optimization model adopts an improved multivariate regression analysis method. By introducing interaction terms and high-order terms, the nonlinear relationship affecting the plate-making parameters is identified. The plate-making parameter optimization model is:
[0106] ;
[0107] in, is the predicted platemaking parameter result, which indicates the platemaking parameter value calculated by the regression model. is the i-th feature, from the feature matrix H after dimensionality reduction, is the regression coefficient, which indicates the influence of each feature and its interaction term on the plate making parameters. is the error term, which is used to represent the prediction error of the model.
[0108] Optionally, the final feature representation The specific optimization goals are:
[0109] ;
[0110] in, represents the F-norm, represents the L1 norm, and is a regularization parameter used to control feature sparsity.
[0111] The beneficial effects of the present invention are:
[0112] The present invention realizes the automatic adjustment and optimization of plate making parameters, improves the fit and comfort of clothing, and significantly improves the optimization efficiency and accuracy of clothing plate making parameters through multi-channel data collection, construction and enhancement processing of comprehensive data models, deep learning and nonlinear relationship extraction, intelligent optimization algorithm and simulation verification, realizes the automation and intelligence of the plate making process, meets the personalized needs of different users, and improves the fit and comfort of clothing. BRIEF DESCRIPTION OF THE DRAWINGS
[0113] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:
[0114] Figure 1 This is a flow chart of a clothing pattern making parameter optimization method based on big data proposed by the present invention. DETAILED DESCRIPTION
[0115] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.
[0116] refer to Figure 1 , a method for optimizing clothing pattern making parameters based on big data, comprising the following steps:
[0117] S1. Obtaining raw data related to clothing pattern making from multiple specific channels;
[0118] S2, using a data fusion algorithm to deeply fuse clothing pattern making parameter sets from different sources to obtain a clothing pattern making parameter set;
[0119] S3, constructing a comprehensive data model through the clothing pattern making parameter set, the comprehensive data model includes standardized body shape data, individual behavior data and environmental factors;
[0120] S4, enhancing the comprehensive data model and generating virtual samples through a generative adversarial network;
[0121] S5. Apply deep learning algorithms to extract patterns and features from the comprehensive data model, identify nonlinear relationships that affect platemaking parameters, and obtain prediction results of the comprehensive data model;
[0122] S6. Based on the prediction results, a dynamic simulation environment is constructed to simulate the effects of different pattern making parameters on the fit and comfort of clothing, allowing users to try on clothing in a virtual environment;
[0123] S7, adjust the plate making parameters according to the feedback from dynamic simulation and virtual fitting, apply the optimized parameters to actual production, and perform secondary optimization by continuously monitoring the feedback data;
[0124] S8. Collect feedback data from the actual production process, including users’ actual wearing experience and quality inspection results of the production line, and iteratively improve the optimization parameters.
[0125] In this implementation, S1 includes the following steps:
[0126] S11. Obtain users’ evaluations on the fit, comfort and style of clothing, and construct a user market feedback dataset M:
[0127] ;
[0128] in, is the i-th market feedback data sample, which includes users’ evaluations on the fit, comfort, and style of clothing;
[0129] S12. Obtain the chest circumference, waist circumference and hip circumference measured by the user, and construct the user body shape dataset U:
[0130] ;
[0131] in, is the body data sample of the jth user, which includes chest circumference, waist circumference and hip circumference;
[0132] S13. Obtain fabric types, process parameters, and quality inspection results for garment production, and construct a garment production dataset:
[0133] ;
[0134] in, is the kth production data sample, which includes fabric type, process parameters and quality inspection results.
[0135] In this implementation, S2 includes the following steps:
[0136] S21, performing data cleaning, deduplication and standardization on the market feedback dataset M, the user body shape dataset U and the production dataset P;
[0137] S22, use the linear discriminant analysis algorithm to deeply fuse the standardized market feedback dataset M', the standardized user body shape dataset U' and the standardized production dataset P'. The linear discriminant analysis algorithm calculates the inter-class scatter matrix between each dataset. and the intra-class scatter matrix , determine the best projection direction and obtain the fused clothing pattern making parameter set F:
[0138] ;
[0139] Among them, LDA represents the linear discriminant analysis process, and outputs the fused clothing pattern making parameter set F.
[0140] In this embodiment, raw data is obtained from multiple channels such as market feedback, user body measurements, and production data, and a data fusion algorithm is used for deep fusion to ensure the comprehensiveness and consistency of the data. The fused data set can reflect the real needs and market trends of different user groups. Compared with the traditional single data source, the present invention greatly improves the diversity and representativeness of the data, reduces the problem of data missing and deviation, and makes the optimization of plate making parameters more reliable and accurate.
[0141] In this implementation, S3 includes the following steps:
[0142] S31, preliminarily processing the fused clothing pattern making parameter set F to extract standardized body shape data, individual behavior data and environmental factors;
[0143] ;
[0144] Each It contains a combination of physical data, behavioral data, and environmental factors;
[0145] S32, constructing the extracted data into a comprehensive data model G;
[0146] S33. Use the improved multimodal clustering algorithm to perform cluster analysis on the standardized body shape data T, identify different body shape groups by combining the geometric characteristics and mechanical properties of the body shape data, and construct a body shape clustering model K:
[0147] ;
[0148] in, Cluster centers representing normalized body size patterns;
[0149] By minimizing the geometric feature error and mechanical property errors The weighted sum of is used for clustering optimization:
[0150] ;
[0151] Among them, the geometric feature error :
[0152] ;
[0153] Mechanical property error :
[0154] ;
[0155] in, and is the weight parameter, represents the i-th body type data sample, represents the mechanical property data of the i-th body type sample, and They represent the geometric characteristics and mechanical properties of the kth cluster center respectively;
[0156] S34, using a behavior feature extraction algorithm to perform pattern recognition on individual behavior data B, combining the time and frequency characteristics of the behavior data, maximizing feature entropy to perform pattern recognition on user behavior feature patterns, and constructing a behavior feature model H:
[0157] ;
[0158] The process of constructing the behavioral feature model is to maximize the feature entropy To optimize:
[0159] ;
[0160] in, Indicates behavioral characteristics The probability distribution of
[0161] S35, using a multimodal fusion algorithm based on graph neural network to fuse the body clustering model K, behavioral feature model H and environmental factor E to construct a comprehensive data model G, and minimize the graph structure loss function Train the comprehensive data model G:
[0162] ;
[0163] Among them, the graph structure loss function Defined by combining node features and edge weights:
[0164] ;
[0165] Where |E| is the number of edges, and Represent the feature representation of nodes u and v respectively, is the global feature mean, is the regularization parameter;
[0166] A comprehensive data model G reflecting the impact of different body shapes, behavioral habits and environmental factors on clothing pattern making parameters is constructed through a multimodal fusion algorithm based on graph neural network.
[0167] In this implementation, the comprehensive data model includes the following components:
[0168] Standardized body shape data T, including the user's chest circumference, waist circumference, hip circumference, shoulder width and height;
[0169] Individual behavior data B, including the user's daily activity level, exercise habits, and clothing preferences;
[0170] Environmental factors E include season, temperature and humidity.
[0171] In this implementation, S4 includes the following steps:
[0172] S41, preprocessing the comprehensive data model G, dividing the data of the comprehensive data model G into a training set and validation set ;
[0173] S42. Construct a multi-level generative adversarial network, which includes a multi-level generator network for generating virtual samples. and a multi-layer discriminator network for distinguishing real samples from virtual samples ;
[0174] S43. Use training set Train a multi-level generative adversarial network to enable the generator to generate realistic virtual samples by minimizing the adaptive loss function of the generator and the multi-level discriminator network. Defined as:
[0175] ;
[0176] in, is the loss function of the generator, represents the expectation of a random noise vector, is the loss weight parameter, represents the feature loss function, which is optimized by calculating the feature difference between virtual samples and real samples;
[0177] Adaptive loss function of the discriminator Defined as:
[0178] ;
[0179] in, is the loss function of the discriminator, represents the expectation of real data, is the loss weight parameter, represents the feature loss function, which is optimized by calculating the feature difference between real samples and virtual samples;
[0180] S44. Through iterative training, use the training set Update the parameters of the generator and discriminator and use the validation set Evaluate model performance until the generator generates virtual samples With real samples Inability to distinguish;
[0181] S45. Construct a feedback loop generative adversarial network based on a multi-level generative adversarial network, and introduce a feedback mechanism from users and experts in the process of generating virtual samples;
[0182] Users and experts evaluate the generated virtual samples in real time through the interface and provide feedback. The feedback loop generative adversarial network dynamically adjusts the parameters of the generator and discriminator based on the feedback.
[0183] User and expert feedback As an adjustment factor, add it to the loss function of the generator and discriminator:
[0184] ;
[0185] ;
[0186] in, represents the expectation of real data, is the feedback adjustment coefficient, which indicates the weight of the feedback on the loss function. represents the expected value of user and expert feedback, represents the expectation of a random noise vector, is the feedback adjustment coefficient, which indicates the weight of the feedback on the loss function. represents the expected value of user and expert feedback, Express expectations of user and expert feedback, user and expert feedback It is expressed as:
[0187] ;
[0188] in, represents the i-th feedback, and N represents the total number of feedbacks;
[0189] S46, the generated virtual sample Add it to the comprehensive data model G, expand the data set G' of the comprehensive data model, and the expanded comprehensive data model G' is expressed as:
[0190] ;
[0191] Among them, G' represents the expanded comprehensive data model, which includes the original comprehensive data model G and the generated virtual samples ;
[0192] S47, post-processing the expanded comprehensive data model G' to remove outliers and noise, and obtain the final comprehensive data model G'' after enhancement processing.
[0193] In this embodiment, a comprehensive data model is constructed, which includes standardized body shape data, individual behavior data and environmental factors. Virtual samples are generated through generative adversarial networks to expand the data set and improve the generalization ability of the model. This model combines multimodal deep learning networks to extract patterns and features and identify nonlinear relationships. By introducing generative adversarial networks, the data sample size is effectively expanded, the generalization ability and robustness of the comprehensive data model are improved, and the influence of different body shapes, behavioral habits and environmental factors on plate making parameters can be more accurately captured.
[0194] In this implementation, the multi-level generator network It consists of multiple levels of fully connected layers, convolutional layers, and attention mechanisms. The input is a random noise vector z and a comprehensive data model G. Virtual samples are generated through a multi-level generation structure. :
[0195] ;
[0196] Among them, z represents a random noise vector, which is the initial input of the generator. represents the generated virtual sample;
[0197] The multi-level generation structure includes a basic generation layer, a high-level generation layer, and a detail generation layer. Each layer is enhanced by skip connections and attention mechanisms. The specific structure is as follows:
[0198] ;
[0199] Among them, x is the input feature map, represents the convolution operation, represents the activation function, Represents the attention mechanism, which is used to enhance feature representation;
[0200] Multi-layer discriminator network It consists of multiple levels of convolutional layers, fully connected layers, and adaptive feature extraction layers, and the input is real samples. and virtual samples , output the distinction result D(G):
[0201] .
[0202] In this implementation, S5 includes the following steps:
[0203] S51. Construct a multimodal deep learning network model, including a multi-layer convolutional neural network and a long short-term memory network, to extract patterns and features from the comprehensive data model and identify nonlinear relationships that affect platemaking parameters;
[0204] S53, multi-layer convolutional neural network extracts the spatial features of the comprehensive data model. The structure includes input layer, multiple convolutional layers, pooling layer and fully connected layer:
[0205] The input layer receives the combined input of standardized body shape data T, individual behavior data B and environmental factors E into the comprehensive data model G''. The convolution layer extracts spatial features through convolution operations. The pooling layer performs feature dimensionality reduction through maximum pooling or average pooling. The fully connected layer maps the output features of the convolution layer and the pooling layer to a high-dimensional space, and the output spatial feature representation ;
[0206] S54, long short-term memory network extracts the time series characteristics of the comprehensive data model, and the structure includes input layer, long short-term memory unit and output layer:
[0207] The input layer receives spatial feature representation As input, the long short-term memory unit extracts time series features through the operation of forget gate, input gate and output gate, and outputs time feature representation ;
[0208] S55. Combine multi-layer convolutional neural network and long short-term memory network, perform feature fusion through hybrid attention mechanism, and generate the final feature representation :
[0209] ;
[0210] in, and is the weight parameter, which is dynamically adjusted through training;
[0211] S56, using the improved non-negative matrix factorization method, combining feature importance and feature sparsity to represent the final feature To reduce the dimension, the goal is to find non-negative matrices W and H such that:
[0212] ;
[0213] Among them, W is the feature weight matrix, H is the feature matrix;
[0214] S57. The selected features are input into the plate-making parameter optimization model. The plate-making parameter optimization model adopts an improved multivariate regression analysis method. By introducing interaction terms and high-order terms, the nonlinear relationship affecting the plate-making parameters is identified. The plate-making parameter optimization model is:
[0215] ;
[0216] in, is the predicted platemaking parameter result, which indicates the platemaking parameter value calculated by the regression model. is the i-th feature, from the feature matrix H after dimensionality reduction, is the regression coefficient, which indicates the influence of each feature and its interaction term on the plate making parameters. is the error term, which is used to represent the prediction error of the model.
[0217] In this embodiment, a deep learning algorithm combining a multi-layer convolutional neural network and a long short-term memory network is used to extract spatial and temporal features from a comprehensive data model, and feature fusion is performed through a hybrid attention mechanism to identify nonlinear relationships that affect platemaking parameters. Compared with the traditional linear regression model, the deep learning model of the present invention can handle complex nonlinear relationships, extract multi-dimensional key features, and improve the accuracy of platemaking parameter optimization, so that the optimization results are more in line with actual production needs and user wearing experience.
[0218] In this embodiment, the final feature representation The specific optimization goals are:
[0219] ;
[0220] in, represents the F-norm, represents the L1 norm, and is a regularization parameter used to control feature sparsity.
[0221] In this implementation, an improved non-negative matrix decomposition method and a multivariate regression analysis method are used to reduce the dimension and optimize the final feature representation of the comprehensive data model. The dynamic simulation environment is used to simulate the effects of different plate-making parameters on the fit and comfort of clothing, and a user and expert feedback mechanism is introduced to adjust the optimization parameters in real time. Through the intelligent optimization algorithm, the repeated trial and error and manual adjustment in the traditional optimization method are reduced, and the optimization efficiency and effect are significantly improved. At the same time, through simulation verification and feedback mechanisms, the feasibility and effectiveness of the optimization parameters in actual production are ensured, and the trial and error costs and production risks are reduced.
[0222] Embodiment 1:
[0223] In modern clothing design and production, the market has higher and higher requirements for clothing fit and comfort. However, the traditional clothing plate making process mainly relies on the designer's experience and manual adjustment, which is not only inefficient, but also difficult to accurately meet the personalized needs of different users. A large clothing manufacturing company decided to apply the method of the present invention to improve the efficiency and accuracy of plate making parameter optimization to meet diversified market needs. The company selected a best-selling women's coat as a pilot, and by collecting market feedback, user body measurements and production data, data fusion and deep learning model training were performed, and finally the plate making parameters were optimized, and simulation verification and actual production were carried out.
[0224] In this embodiment, how to apply the clothing pattern making parameter optimization method based on big data in the above scenario will be introduced in detail, and the feasibility and effectiveness of the present invention will be demonstrated through specific training samples.
[0225] First, we obtain raw data related to the pattern making of this women's coat from multiple specific channels. These channels include:
[0226] Market feedback data: collected user reviews of 5,000 coats sold in the past year, including indicators such as fit, comfort and style satisfaction.
[0227] User body shape data: Detailed body shape measurements were taken on 1,000 target customers, recording parameters such as chest circumference, waist circumference, hip circumference, shoulder width and height.
[0228] Production data: Production data including fabric types, process parameters and quality inspection results were obtained from the production line, totaling 3,000 records.
[0229] The collected data is cleaned, deduplicated and standardized, and the clothing pattern making parameter sets from different sources are deeply integrated using the linear discriminant analysis algorithm to obtain standardized market feedback data sets, standardized user body shape data sets and standardized production data sets. Then, a comprehensive data model is constructed.
[0230] The comprehensive data model is enhanced using a generative adversarial network, which expands the data set by generating virtual samples and improves the generalization ability of the model. A multimodal deep learning network model is constructed, including a multi-layer convolutional neural network and a long short-term memory network, to extract patterns and features from the comprehensive data model and identify the nonlinear relationships that affect the platemaking parameters. After data preprocessing and enhancement, a deep learning model is constructed. The multi-layer convolutional neural network extracts spatial features, and the long short-term memory network extracts time series features. The hybrid attention mechanism is used to perform feature fusion and generate the final feature representation. The improved non-negative matrix decomposition method is used to reduce the dimension of the final feature representation. The multivariate regression analysis method is used to introduce interaction terms and high-order terms to identify the nonlinear relationships that affect the platemaking parameters and obtain an optimized platemaking parameter model.
[0231] The optimized pattern-making parameters are applied to the simulation environment to simulate the effect of different pattern-making parameters on the fit and comfort of the garment. Through the virtual fitting system, users are allowed to try on the optimized jackets in a virtual environment, collect user feedback, and further adjust the pattern-making parameters. Finally, the optimized parameters are used for small-batch production, and quality inspections and user fitting feedback are carried out to ensure the feasibility and effectiveness of the optimized parameters in actual production.
[0232] In order to verify the effectiveness of the method of the present invention, we compared the method of the present invention with the traditional method, and the specific data are shown in Table 1 below:
[0233] Table 1 Comparison data between the method of the present invention and the traditional method
[0234] project Traditional methods Method of the present invention Data collection channels Single channel Multi-channel integration Data processing time 10 days 3 days Data fusion and standardization accuracy 70% 95% Model training time 7 days 4 days Model prediction accuracy 85% 98% Plate making parameter optimization efficiency Low high Simulation verification and actual fitting times 10 times 2 times User satisfaction (fit) 80% 95% User satisfaction (comfort) 75% 92% Trial and error costs high Low Total production cycle 30 days 20 days
[0235] It can be seen from the above data that the method of the present invention is superior to traditional methods in data processing, model training, prediction accuracy, optimization efficiency, etc., which significantly improves the efficiency and accuracy of platemaking parameter optimization and reduces trial and error costs and production cycles.
[0236] In order to further demonstrate the feasibility of the method of the present invention, we selected a specific training sample for detailed description.
[0237] Training sample description: User A, female, 30 years old, height 165cm, chest circumference 88cm, waist circumference 70cm, hip circumference 95cm, shoulder width 40cm. The user's daily activities are mainly office work, occasionally outdoor sports, and the requirements for clothing are fit, comfortable and fashionable.
[0238] Data Collection:
[0239] Market feedback data: Collect reviews of similar coats purchased by user A, including a fit score of 4.5 points (out of 5 points), a comfort score of 4.2 points, and a style satisfaction score of 4.7 points.
[0240] User body data: Through body measurement, the body data of user A is recorded (chest 88cm, waist 70cm, hip 95cm, shoulder width 40cm, height 165cm).
[0241] Production data: Obtain the type of fabric used, process parameters and quality test results. The fabric is wool blended, the process parameters are standard, and the quality test results are qualified.
[0242] Simulation verification and feedback adjustment:
[0243] In the simulation environment, user A was simulated to try on the optimized jacket and feedback from user A was collected. The fit score was improved to 4.8 and the comfort score was improved to 4.6. The pattern making parameters were further adjusted based on the feedback, and small batch production and quality inspection were carried out.
[0244] Through the above steps, the optimized plate-making parameters are finally obtained, and the produced coat is highly praised by user A, and the fit and comfort are significantly improved. Through the application of the method of the present invention, not only the efficiency and accuracy of plate-making parameter optimization are improved, but also the trial and error cost and production cycle are effectively reduced, which fully verifies the feasibility and effectiveness of the method of the present invention.
[0245] The present invention realizes the automatic adjustment and optimization of plate making parameters, improves the fit and comfort of clothing, and significantly improves the optimization efficiency and accuracy of clothing plate making parameters through multi-channel data collection, construction and enhancement processing of comprehensive data models, deep learning and nonlinear relationship extraction, intelligent optimization algorithm and simulation verification, realizes the automation and intelligence of the plate making process, meets the personalized needs of different users, and improves the fit and comfort of clothing.
[0246] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
Claims
1. A method for optimizing clothing plate making parameters based on big data, characterized in that: The steps include: S1. Obtain raw data related to clothing pattern making from multiple specific channels; S2, using a data fusion algorithm to deeply fuse clothing pattern making parameter sets from different sources to obtain a clothing pattern making parameter set; S3, constructing a comprehensive data model through the clothing pattern making parameter set, the comprehensive data model includes standardized body shape data, individual behavior data and environmental factors; S4, enhancing the comprehensive data model and generating virtual samples through a generative adversarial network; S5. Apply deep learning algorithms to extract patterns and features from the comprehensive data model, identify nonlinear relationships that affect platemaking parameters, and obtain prediction results of the comprehensive data model; S6. Based on the prediction results, a dynamic simulation environment is constructed to simulate the effects of different pattern making parameters on the fit and comfort of clothing, allowing users to try on clothing in a virtual environment; S7, adjust the plate making parameters according to the feedback from dynamic simulation and virtual fitting, apply the optimized parameters to actual production, and perform secondary optimization by continuously monitoring the feedback data; S8. Collect feedback data from the actual production process, including users’ actual wearing experience and quality inspection results of the production line, and iteratively improve the optimization parameters.
2. The method for optimizing clothing plate making parameters based on big data according to claim 1, characterized in that: The S1 comprises the following steps: S11. Obtain users’ evaluations on the fit, comfort and style of clothing, and construct a user market feedback dataset M: ; in, is the i-th market feedback data sample, which includes users’ evaluations on the fit, comfort, and style of clothing; S12. Obtain the chest circumference, waist circumference and hip circumference measured by the user, and construct the user body shape dataset U: ; in, is the body data sample of the jth user, which includes chest circumference, waist circumference and hip circumference; S13. Obtain fabric types, process parameters, and quality inspection results for garment production, and construct a garment production dataset: ; in, is the kth production data sample, which includes fabric type, process parameters and quality inspection results.
3. The method for optimizing clothing plate making parameters based on big data according to claim 2, characterized in that: The S2 comprises the following steps: S21, performing data cleaning, deduplication and standardization on the market feedback dataset M, the user body shape dataset U and the production dataset P; S22, use the linear discriminant analysis algorithm to deeply fuse the standardized market feedback dataset M', the standardized user body shape dataset U' and the standardized production dataset P'. The linear discriminant analysis algorithm calculates the inter-class scatter matrix between each dataset. and the intra-class scatter matrix , determine the best projection direction and obtain the fused clothing pattern making parameter set F: ; Among them, LDA represents the linear discriminant analysis process, and outputs the fused clothing pattern making parameter set F.
4. The method for optimizing clothing plate making parameters based on big data according to claim 3, characterized in that: The S3 comprises the following steps: S31, preliminarily processing the fused clothing pattern making parameter set F to extract standardized body shape data, individual behavior data and environmental factors; ; Each It contains a combination of physical data, behavioral data, and environmental factors; S32, constructing the extracted data into a comprehensive data model G; S33. Use the improved multimodal clustering algorithm to perform cluster analysis on the standardized body shape data T, identify different body shape groups by combining the geometric characteristics and mechanical properties of the body shape data, and construct a body shape clustering model K: ; in, Cluster centers representing normalized body size patterns; By minimizing the geometric feature error and mechanical property errors The weighted sum of is used for clustering optimization: ; Among them, the geometric feature error : ; Mechanical property error : ; in, and is the weight parameter, represents the i-th body type data sample, represents the mechanical property data of the i-th body type sample, and They represent the geometric characteristics and mechanical properties of the kth cluster center respectively; S34, using a behavior feature extraction algorithm to perform pattern recognition on individual behavior data B, combining the time and frequency characteristics of the behavior data, maximizing feature entropy to perform pattern recognition on user behavior feature patterns, and constructing a behavior feature model H: ; The process of constructing the behavioral feature model is to maximize the feature entropy To optimize: ; in, Indicates behavioral characteristics The probability distribution of S35, using a multimodal fusion algorithm based on graph neural network to fuse the body clustering model K, behavioral feature model H and environmental factor E to construct a comprehensive data model G, and minimize the graph structure loss function Train the comprehensive data model G: ; Among them, the graph structure loss function Defined by combining node features and edge weights: ; Where |E| is the number of edges, and Represent the feature representation of nodes u and v respectively, is the global feature mean, is the regularization parameter; A comprehensive data model G reflecting the impact of different body shapes, behavioral habits and environmental factors on clothing pattern making parameters is constructed through a multimodal fusion algorithm based on graph neural network.
5. The method for optimizing clothing plate making parameters based on big data according to claim 4, characterized in that: The comprehensive data model includes the following components: Standardized body shape data T, including the user's chest circumference, waist circumference, hip circumference, shoulder width and height; Individual behavior data B, including the user's daily activity level, exercise habits, and clothing preferences; Environmental factors E include season, temperature and humidity.
6. The method for optimizing clothing plate making parameters based on big data according to claim 4, characterized in that: The S4 comprises the following steps: S41, preprocessing the comprehensive data model G, dividing the data of the comprehensive data model G into a training set and validation set ; S42. Construct a multi-level generative adversarial network, which includes a multi-level generator network for generating virtual samples. and a multi-layer discriminator network for distinguishing real samples from virtual samples ; S43. Use training set Train a multi-level generative adversarial network to enable the generator to generate realistic virtual samples by minimizing the adaptive loss function of the generator and the multi-level discriminator network. Defined as: ; in, is the loss function of the generator, represents the expectation of a random noise vector, is the loss weight parameter, represents the feature loss function, which is optimized by calculating the feature difference between virtual samples and real samples; Adaptive loss function of the discriminator Defined as: ; in, is the loss function of the discriminator, represents the expectation of real data, is the loss weight parameter, represents the feature loss function, which is optimized by calculating the feature difference between real samples and virtual samples; S44. Through iterative training, use the training set Update the parameters of the generator and discriminator and use the validation set Evaluate model performance until the generator generates virtual samples With real samples Inability to distinguish; S45. Construct a feedback loop generative adversarial network based on a multi-level generative adversarial network, and introduce a feedback mechanism from users and experts in the process of generating virtual samples; Users and experts evaluate the generated virtual samples in real time through the interface and provide feedback. The feedback loop generative adversarial network dynamically adjusts the parameters of the generator and discriminator based on the feedback. User and expert feedback As an adjustment factor, add it to the loss function of the generator and discriminator: ; ; in, represents the expectation of real data, is the feedback adjustment coefficient, which indicates the weight of the feedback on the loss function. represents the expected value of user and expert feedback, represents the expectation of a random noise vector, is the feedback adjustment coefficient, which indicates the weight of the feedback on the loss function. represents the expected value of user and expert feedback, Express expectations of user and expert feedback, user and expert feedback It is expressed as: ; in, represents the i-th feedback, and N represents the total number of feedbacks; S46, the generated virtual sample Add it to the comprehensive data model G, expand the data set G' of the comprehensive data model, and the expanded comprehensive data model G' is expressed as: ; Among them, G' represents the expanded comprehensive data model, which includes the original comprehensive data model G and the generated virtual samples ; S47, post-processing the expanded comprehensive data model G' to remove outliers and noise, and obtain the final comprehensive data model G'' after enhancement processing.
7. The method for optimizing clothing pattern making parameters based on big data according to claim 6, characterized in that: The multi-level generator network It consists of multiple levels of fully connected layers, convolutional layers, and attention mechanisms. The input is a random noise vector z and a comprehensive data model G. Virtual samples are generated through a multi-level generation structure. : ; Among them, z represents a random noise vector, which is the initial input of the generator. represents the generated virtual sample; The multi-level generation structure includes a basic generation layer, a high-level generation layer, and a detail generation layer. Each layer is enhanced by skip connections and attention mechanisms. The specific structure is as follows: ; Among them, x is the input feature map, represents the convolution operation, represents the activation function, Represents the attention mechanism, which is used to enhance feature representation; Multi-layer discriminator network It consists of multiple levels of convolutional layers, fully connected layers, and adaptive feature extraction layers, and the input is real samples. and virtual samples , output the distinction result D(G): 。 8. The method for optimizing clothing plate making parameters based on big data according to claim 6, characterized in that: The S5 comprises the following steps: S51. Construct a multimodal deep learning network model, including a multi-layer convolutional neural network and a long short-term memory network, to extract patterns and features from the comprehensive data model and identify nonlinear relationships that affect platemaking parameters; S53, multi-layer convolutional neural network extracts the spatial features of the comprehensive data model. The structure includes input layer, multiple convolutional layers, pooling layer and fully connected layer: The input layer receives the combined input of standardized body shape data T, individual behavior data B and environmental factors E into the comprehensive data model G''. The convolution layer extracts spatial features through convolution operations. The pooling layer performs feature dimensionality reduction through maximum pooling or average pooling. The fully connected layer maps the output features of the convolution layer and the pooling layer to a high-dimensional space, and the output spatial feature representation ; S54, long short-term memory network extracts the time series characteristics of the comprehensive data model, and the structure includes input layer, long short-term memory unit and output layer: The input layer receives spatial feature representation As input, the long short-term memory unit extracts time series features through the operation of forget gate, input gate and output gate, and outputs time feature representation ; S55. Combine multi-layer convolutional neural network and long short-term memory network, perform feature fusion through hybrid attention mechanism, and generate the final feature representation : ; in, and is the weight parameter, which is dynamically adjusted through training; S56, using the improved non-negative matrix factorization method, combining feature importance and feature sparsity to represent the final feature To reduce the dimension, the goal is to find non-negative matrices W and H such that: ; Among them, W is the feature weight matrix, H is the feature matrix; S57. The selected features are input into the plate-making parameter optimization model. The plate-making parameter optimization model adopts an improved multivariate regression analysis method. By introducing interaction terms and high-order terms, the nonlinear relationship affecting the plate-making parameters is identified. The plate-making parameter optimization model is: ; in, is the predicted platemaking parameter result, which indicates the platemaking parameter value calculated by the regression model. is the i-th feature, from the feature matrix H after dimensionality reduction, is the regression coefficient, which indicates the influence of each feature and its interaction term on the plate making parameters. is the error term, which is used to represent the prediction error of the model.
9. The method for optimizing clothing pattern making parameters based on big data according to claim 8, characterized in that: The final feature representation The specific optimization goals are: ; in, represents the F-norm, represents the L1 norm, and is a regularization parameter used to control feature sparsity.
