National costume design method based on deep learning
Through the ethnic clothing design method based on deep learning, the problems of traditional design taking time, difficulty and low data processing efficiency are solved, and the rapid generation of design sketches and innovative design solutions are achieved, which enhances the embodiment of national culture and market adaptability.
Patent Information
- Application Number
- CN202411588284.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-07
- Publication Date
- 2025-02-28
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional ethnic clothing design takes time and is difficult to design, and it is difficult to meet the rapidly changing market demand and the update speed of fashion trends. In addition, existing digital design tools have limitations in processing data related to ethnic clothing, and are inefficient in sorting and classification and are prone to errors.
The ethnic clothing design method based on deep learning is adopted, and ethnic clothing data is collected through Internet crawlers. Combined with field investigations and cooperation between ethnic culture research institutions, data annotation and preprocessing is carried out, and a deep learning model is built to learn the characteristics and styles of ethnic clothing and generate preliminary design sketches.
It has achieved rapid generation of design sketches, improved design efficiency, enhanced the innovation and systematicity of the design, and can more accurately reflect the connotation of national culture and meet the rapid changes in market demand.
Smart Images
Figure CN119538348A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of clothing design, and in particular to a method for designing ethnic clothing based on deep learning. Background Art
[0002] Traditional ethnic clothing design mainly relies on the personal experience and manual drawing of designers. The design process is time-consuming and difficult to respond quickly to changes in market demand. At the same time, since it is difficult for designers to quickly and deeply understand the images and symbols of the ethnic culture they design during work, designers need to spend a lot of time on sketching and repeated revisions. It often takes a long time from inspiration to the final design plan. The excavation and application of ethnic cultural elements are also relatively limited, and there is a lack of innovative and systematic methods to integrate modern design concepts with ethnic characteristics. With the advancement of modern technology, designers have more sources of ethnic clothing elements. The design schemes that explore traditional ethnic clothing culture and integrate it into modern clothing are widely welcomed by consumers. With the huge source of data obtained, the traditional reliance on designers to handle clothing types, patterns, and colors separately has difficulties, such as:
[0003] In the process of designing and selecting complex ethnic costumes, designers need to spend a lot of time looking up ethnic culture and symbolic meanings, which takes longer and is more difficult to design. It is difficult to meet the rapidly changing market demand and the speed of fashion trends. Traditional design patterns are relatively fixed, and designers are easily constrained by existing design styles and ways of thinking. In addition, some existing digital design tools have limitations when processing ethnic costume-related data. After obtaining a large amount of ethnic costume images, patterns, colors and other data, designers need to classify them according to the inspection materials, which has a high error rate, low overall efficiency and is prone to errors.
[0004] In view of the above problems, it is urgent to carry out innovative designs based on the original national costume design system. Summary of the invention
[0005] The purpose of the present invention is to provide a method for designing ethnic clothing based on deep learning, so as to solve the problems raised in the above background technology that existing designers need longer time to design ethnic clothing, the design is more difficult, and it is difficult to meet the rapidly changing market demand and the updating speed of fashion trends; and some existing digital design tools have limitations in processing ethnic clothing related data, and the organization and classification of clothing related data are inefficient and prone to errors.
[0006] To achieve the above object, the present invention provides the following technical solution: a method for designing ethnic clothing based on deep learning, comprising the following steps:
[0007] S1. Use Internet crawlers to collect images, patterns, colors and other data of various ethnic costumes on the network platform, including photos of traditional costumes, illustrations of costumes in historical documents, etc., cooperate with photos taken in ethnic areas during field investigations, and cooperate with ethnic cultural research institutions and scholars to ensure the accuracy and professionalism of the data obtained;
[0008] S2. Organize a professional design team and ethnic culture experts to annotate the collected data, including ethnic categories, clothing styles, pattern types, color matching, craft characteristics, cultural implications, etc., and specify detailed annotation specifications and standards to ensure the consistency and accuracy of the annotation;
[0009] S3. Perform image denoising, image enhancement, image segmentation and other pre-processing operations on the annotated image data through image recognition technology to improve image quality, clean and organize the acquired text data, remove irrelevant and redundant information, and convert it into a structured data format to associate and integrate with the image data;
[0010] S4. Select TensorFlow as the deep learning framework, combine the microbial agent neural network with multiple convolutional layers, pooling layers and fully connected layers, and continuously adjust the parameters and optimize the algorithm to build a deep learning model. Combine the residual connection and global average pooling layer to improve the accuracy and generalization ability of the model. After daily training, the model is optimized by the algorithm to facilitate learning the characteristics and style of national costumes;
[0011] S5. By establishing an interactive interface, users can associate clothing types, ethnic style preferences, color matching requirements and pattern element selection according to design requirements. After the user describes the design details, after text parsing and processing, it is converted into information coding and data signals that can be understood by the model, and a preliminary sketch is generated through comprehensive analysis and processing by the deep learning model;
[0012] S6. After obtaining the design sketches produced in S5, the designer evaluates the clothing styles, further optimizes some details, and forms a complete ethnic clothing design plan in conjunction with the annotated ethnic cultural information.
[0013] By adopting the above technical solution and utilizing the deep learning model, a preliminary design sketch can be quickly generated based on the input design requirements and related parameters.
[0014] Preferably, the data such as images, patterns, colors, etc. of ethnic costumes obtained in S1 are the main source of information for establishing an ethnic culture database, and are obtained from various museums and ethnic culture research institutions at home and abroad, field photography and research in ethnic concentrated areas, and the use of web crawler technology to screen and select appropriate ethnic costume data.
[0015] By adopting the above technical solutions, we can enrich the information and data sources of ethnic costumes and fully obtain most of the clothing design units.
[0016] Preferably, during the process of labeling the acquired data, S2 simultaneously assists in establishing a national culture database, and after labeling all acquired data, the labeled data are classified and sorted.
[0017] The above technical solution is adopted to provide more precise standard division and option processing for the characteristics of national costumes.
[0018] Preferably, the ethnic culture database is classified in such a way that clothing data belonging to the same ethnic group are grouped together to form a sub-data set based on ethnicity. In each ethnic sub-data set, the data is further subdivided according to clothing styles. A separate pattern database is established in the ethnic culture database to classify all marked patterns according to type.
[0019] By adopting the above technical solution, the collected national cultural elements can comprehensively include detailed information of various national cultural elements.
[0020] Preferably, the colors of the clothing data in the ethnic culture database are classified according to the color hue, brightness, saturation and other characteristics, and the clothing craftsmanship is classified according to different production processes, and the data in the ethnic culture database is assisted by data annotation by a professional team.
[0021] By adopting the above technical solution, the establishment of the database combined with the deep learning algorithm model ensures that these elements are accurately applied to clothing, so that the designed clothing can more truly reflect the national cultural connotation.
[0022] Preferably, the professional team includes experts and clothing designers who have in-depth research on the history, cultural implications, traditional customs, etc. of ethnic costumes and are familiar with different ethnic cultures, and classify and label the ethnic types and styles of clothing.
[0023] The above technical solution is adopted to facilitate providing more detailed information content to assist in the excavation of traditional ethnic culture.
[0024] Preferably, the professional team uses the research results of ethnology and official classifications to determine the typical characteristics of each ethnic costume as a basis for judgment, formulates clear ethnic classification standards for ethnic category labeling, clarifies the labeling categories and standards, and establishes an audit mechanism to review the labeling.
[0025] By adopting the above technical solution, the acquired source of ethnic clothing characteristics has a more accurate classification source.
[0026] Preferably, the convolution layer in the deep learning algorithm model in S4 has multiple convolution kernels of different scales, the convolution layer adopts a depth-separable convolution structure, and the residual information between the convolution layers is processed by the joint residual connection, and the output layer in the deep learning algorithm model adopts a softmax activation function to output the probability distribution of different clothing style categories.
[0027] By adopting the above technical solution, all clothing types in the database are pulled through the deep learning algorithm model, and novel combination schemes for ethnic clothing design are added.
[0028] Preferably, the model training of the deep learning algorithm model relies on providing learning materials for 70% training sets, 20% validation sets and 10% test sets divided in the national culture database, and selects the cross entropy loss function to provide image prediction result training for clothing style classification.
[0029] The above technical solution is adopted to enrich the processing of image fitting in the deep learning algorithm model training process.
[0030] Preferably, the cross entropy loss function uses an Adam adaptive learning rate optimization algorithm to balance model training parameters during training, and adds performance indicators to monitor the number of training rounds on the observed validation set.
[0031] The above technical solutions are adopted to improve the protection of deep learning algorithm model training and avoid overfitting.
[0032] Compared with the prior art, the beneficial effects of the present invention are: the ethnic clothing design method based on deep learning:
[0033] 1. Through web crawlers, we extract information about ethnic costumes in modern clothing design, and combine it with clothing images and videos from offline field surveys, as well as clothing images and text descriptions in historical data from the Institute of Ethnology. After processing the image data, we can quickly extract features of patterns and clothing forms in the images. After encoding the text data, we can associate the annotated text data with the image data, and thus establish a comprehensive database of ethnic cultural elements, providing rich materials and sources of inspiration for design. By converting text descriptions into digital vectors, we can easily input image data into the deep learning model for learning and processing.
[0034] 2. Using the deep learning model, it is possible to quickly generate preliminary design sketches based on the input design requirements and related parameters. After learning and analyzing a large number of ethnic clothing samples in the ethnic culture database, the deep learning model conducts innovative design based on the set needs and combines the potential patterns and rules in the data to generate some unique design elements and combinations that designers may not have thought of. At the same time, the rich and detailed information in the database makes the designed clothing more realistically reflect the connotation of ethnic culture. By combining ethnic cultural elements with modern clothing design, innovative design solutions that integrate multiple elements and styles are generated, so that more people will be interested in the culture of the nation, thereby promoting the promotion and development of national culture;
[0035] 3. This design method allows users to participate in the design process based on specific needs. As the deep learning model learns and analyzes large amounts of data, it can more accurately grasp the design points and style characteristics of ethnic costumes. Then, based on customer requirements, while maintaining ethnic characteristics, it adjusts the style and details of clothing to make it more suitable for daily activities while reflecting the customer's personal style. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Figure 1 It is a schematic diagram of the steps of the ethnic clothing design process of the present invention;
[0037] Figure 2 This is a schematic diagram of the auxiliary process of labeling ethnic clothing image data of the present invention;
[0038] Figure 3 It is a detailed schematic diagram of the process of designing ethnic clothing of the present invention. DETAILED DESCRIPTION
[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0040] See also Figure 1-Figure 3 The present invention provides a technical solution: a method for designing ethnic clothing based on deep learning, comprising the following steps:
[0041] S1. Use Internet crawlers to collect images, patterns, colors and other data of various ethnic costumes on the network platform, including photos of traditional costumes, illustrations of costumes in historical documents, etc., cooperate with photos taken in ethnic areas during field investigations, and cooperate with ethnic cultural research institutions and scholars to ensure the accuracy and professionalism of the data obtained;
[0042] The data of images, patterns, colors, etc. of ethnic costumes obtained in S1 are the main sources of information for establishing the ethnic culture database. They are obtained from various museums and ethnic culture research institutions at home and abroad, field photography and research in ethnic concentrated areas, and the use of web crawler technology to screen and select appropriate ethnic costume data. The ethnic culture database is classified, and the clothing data belonging to the same ethnic group is classified into one category to form a sub-dataset based on ethnicity. In each ethnic sub-dataset, it is further subdivided according to clothing styles. In addition, a separate pattern database is established in the ethnic culture database to sort all the marked patterns according to type. The colors of the clothing data in the ethnic culture database are classified according to color. The hue, brightness, saturation and other characteristics of the colors are classified, and the clothing crafts are classified according to different production processes. The data in the ethnic culture database is assisted by data annotation by a professional team. The professional team has experts and clothing designers who have in-depth research on the history, cultural implications, traditional customs, etc. of ethnic costumes and are familiar with different ethnic cultures. They classify and annotate the ethnic types and styles of clothing. The professional team uses the research results of ethnology and official classifications to determine the typical characteristics of each ethnic costume as the basis for judgment. For the ethnic category annotation, clear ethnic classification standards are formulated, and the annotation categories and standards are clarified. At the same time, an audit mechanism is established to review the annotations;
[0043] S2. Organize a professional design team and ethnic culture experts to annotate the collected data, including ethnic categories, clothing styles, pattern types, color matching, craft characteristics, cultural implications, etc., and specify detailed annotation specifications and standards to ensure the consistency and accuracy of the annotation;
[0044] S2 assists in the establishment of the national culture database while labeling the acquired data. After labeling all acquired data, the labeled data are classified and sorted.
[0045] S3. Perform image denoising, image enhancement, image segmentation and other pre-processing operations on the annotated image data through image recognition technology to improve image quality, clean and organize the acquired text data, remove irrelevant and redundant information, and convert it into a structured data format to associate and integrate with the image data;
[0046] S4. Select TensorFlow as the deep learning framework, combine the microbial agent neural network with multiple convolutional layers, pooling layers and fully connected layers, and continuously adjust the parameters and optimize the algorithm to build a deep learning model. Combine the residual connection and global average pooling layer to improve the accuracy and generalization ability of the model. After daily training, the model is optimized by the algorithm to facilitate learning the characteristics and style of national costumes;
[0047] The convolution layer in the deep learning algorithm model in S4 has multiple convolution kernels of different scales. The convolution layer adopts a depth-separable convolution structure and a residual connection is used to process the residual information between the convolution layers. The output layer in the deep learning algorithm model uses a softmax activation function to output the probability distribution of different clothing style categories.
[0048] S5. By establishing an interactive interface, users can associate clothing types, ethnic style preferences, color matching requirements and pattern element selection according to design requirements. After the user describes the design details, after text parsing and processing, it is converted into information coding and data signals that can be understood by the model, and a preliminary sketch is generated through comprehensive analysis and processing by the deep learning model;
[0049] The model training of the deep learning algorithm model relies on the 70% training set, 20% validation set and 10% test set divided in the national culture database to provide learning materials, and the cross entropy loss function is used to provide image prediction result training for clothing style classification. The cross entropy loss function uses the Adam adaptive learning rate optimization algorithm to balance the model training parameters during the training process, and adds performance indicators to monitor the number of training rounds on the observed validation set;
[0050] S6. After obtaining the design sketches produced in S5, the designer evaluates the clothing styles, further optimizes some details, and forms a complete ethnic clothing design plan in conjunction with the annotated ethnic cultural information.
[0051] In conjunction with the accompanying drawings Figure 1-Figure 3 As shown in Figure 1, S1 uses image data obtained by web crawlers and cooperates with various museums and ethnic culture research institutions at home and abroad to provide high-definition images of ethnic costumes in their collections, scanned copies of relevant historical documents, etc., and then organizes a professional team to go deep into ethnic areas for field shooting and research, recording the traditional costumes worn by local people in daily life, the costumes worn on special festivals or ceremonies, and the process of traditional craftsmen making costumes, etc. During the shooting, local cultural scholars and elders were interviewed simultaneously to collect oral materials such as cultural stories, meanings, and production processes behind the costumes, and to organize and record them, so that the ethnic culture database can provide documentary information in the data sorting process. After the data is acquired, image preprocessing is performed, and image filtering algorithms such as median filtering and Gaussian filtering are used to remove noise points and impurities in the image, improve the clarity and quality of the image, and adjust the image contrast, brightness, saturation and other parameters to enhance the details and color expression of the clothing in the image. After that, the size and resolution of all images are normalized to facilitate the input and calculation of the model. At the same time, the acquired text data is encoded using unique hot encoding or word embedding technology, and the text data is converted into digital vector form, so that it can be input into the deep learning model together with the image data for learning and processing;
[0052] In conjunction with the accompanying drawings Figure 2 As shown, by forming a professional annotation team, including ethnic culture experts, fashion designers and data annotation personnel, ethnic culture experts mark the ethnic group to which each ethnic costume belongs. For some clothes with mixed styles or unclear regional characteristics, accurate judgment and annotation are made by consulting relevant literature, consulting experts, etc. Fashion designers are used to divide the style types of clothing in detail and further subdivide their style characteristics. At the same time, the patterns on the clothing are carefully annotated, including the name of the pattern, the type of pattern, the location of the pattern and the meaning of the pattern. When using professional color recognition tools, the main color and auxiliary color of the clothing are accurately annotated, and its symbolic meaning in the ethnic culture is recorded. The production process used in each piece of clothing is indicated, and the characteristics and difficulty level of the process are described. After that, it is classified and sorted according to ethnicity, clothing type, pattern, color and process, and multiple sub-data sets are obtained and summarized in the ethnic culture database. When the model generates the corresponding clothing design draft pattern, the corresponding clothing type, pattern, color and process annotation are provided. The review mechanism established during the annotation period is used to improve the accuracy of the annotation content of the ethnic clothing pattern;
[0053] The construction of the deep learning model uses convolutional neural network as the basic model architecture, and introduces multiple convolution kernels of different scales to capture features of different sizes and levels in clothing images. Among them, smaller convolution kernels can focus on local detail features, such as pattern texture and lines, and larger convolution kernels can obtain more macro features, such as the overall shape of clothing, style structure, etc. By combining convolution layers of different scales, the model can comprehensively learn various features of national costumes. In order to reduce the number of parameters and calculation amount of the model and improve the operation efficiency of the model, a deep separable convolution structure is used in some convolution layers. This structure decomposes the traditional convolution operation into two steps: deep convolution and point-by-point convolution. Without significantly reducing the performance of the model, the complexity of the model is greatly reduced.
[0054] Add residual connections to the model to solve the gradient vanishing problem that may occur as the number of network layers increases, so that the model can better learn deep features. The residual connection directly adds the input to the output after processing by multiple convolutional layers, so that the network can more easily learn the residual information between the input and output, thereby improving the training effect and generalization ability of the model. In the last few layers of the model, the global average pooling layer is used instead of the traditional fully connected layer. The global average pooling operation is performed on each feature map to convert it into a feature point, which reduces the number of model parameters and the risk of overfitting. At the same time, it can better retain the spatial information of the image. According to the requirements of the design task, design a suitable output layer;
[0055] The data in the ethnic culture database is divided into a training set, a validation set, and a test set. The data in the training set is randomly rotated, randomly scaled, randomly cropped, and horizontally flipped to provide a technology for the model to learn the characteristics of ethnic clothing at different angles, sizes, and postures. When classifying clothing styles, the cross entropy loss function is used to measure the difference between the probability distribution predicted by the model and the true label. The model parameters are optimized by minimizing the cross entropy loss to make the model's prediction results as close to the true label as possible. For the image generation task, the mean square error loss function or the perceptual loss function can be used to calculate the difference in pixel values between the generated image and the target image. The perceptual loss function takes into account the difference in the perception of the human visual system to the image. The model is optimized by comparing the difference between the generated image and the target image in the high-level feature space, so that the generated image is visually closer to the real ethnic clothing image. By using an optimization algorithm with an adaptive learning rate, the learning rate is automatically adjusted according to the gradient information during the model training process, so that the model can converge quickly in the early stage of training and can more stably approach the optimal solution in the later stage of training.
[0056] When using, determine the appropriate batch size based on the computer hardware resources and data size of the model. A larger batch size can fully utilize the parallel computing capabilities of the GPU and increase the training speed, but it may lead to a decrease in the generalization ability of the model.
[0057] A smaller batch size allows the model to learn the features of each sample more finely, but the training speed will be relatively slow. In practice, we usually try different batch sizes to find a balance point and determine the appropriate number of training rounds. Too few training rounds may cause the model to fail to fully learn the features in the data, and too many training rounds may cause overfitting. Generally, the optimal number of training rounds is determined by observing the performance indicators on the validation set during the training process.
[0058] After the model training is completed, the interactive interface is connected, and user needs are input, such as clothing type selection, style preference setting, color matching requirements, pattern element selection, and some design details, as well as fabric material and special process treatment. The system through the interactive interface parses and processes the user's input needs, and converts them into a form that the model can understand, so that the model enters the ethnic cultural clothing database to retrieve corresponding data and filter related elements. In the clothing type classification data set, clothing style examples that match the clothing type selected by the user are found. According to the style preferences set by the user, clothing samples and elements with corresponding style characteristics are screened out from the database. According to the color matching requirements specified by the user, similar or matching color combinations are found in the color classification database. According to the pattern elements selected by the user, the corresponding pattern information is extracted from the pattern database. Then, the retrieved relevant clothing style features, color features, pattern features and other elements and user demand information are input into the trained deep learning model. The model conducts comprehensive analysis and processing of these information, based on Based on the fused feature vector, the model gradually generates design sketches through a series of convolution, deconvolution and upsampling operations. During the generation process, the model will fill and optimize the sketches in detail according to the learned style and design rules of ethnic costumes, thereby generating multiple preliminary design sketches. By introducing random factors into the model, some parameters are randomly perturbed or different initialization conditions are adopted in the generation process, so that the sketches generated by the model each time have certain differences. By displaying multiple generated sketches on the interactive interface, it is convenient for designers to evaluate and modify them. After the overall evaluation and modification of the sketches, the designer will deeply optimize some details as needed to meet the personalized aesthetic needs of users, and obtain clothing with national characteristics and in line with their own personality in accordance with user needs. After the design scheme is more diversified and personalized, it can meet the needs of different users, enhance the application of national cultural elements in clothing design, inherit and promote national culture, improve the market competitiveness of enterprises, and provide new technical support for the development of the national clothing industry.
[0059] Working principle: When using this ethnic costume design method based on deep learning, first set up a special data collection team, cooperate with ethnic culture research institutions, museums, etc. to obtain authoritative ethnic costume data, and use web crawler technology to obtain relevant ethnic costume pictures and information. Then, clean and screen the collected data, remove unqualified data, and annotate and organize them according to the classification standards predetermined by the expert team to ensure data accuracy and consistency. The organized and annotated data enters the ethnic culture database, and then divides the organized training set, validation set, and test set. The model is built through the convolutional neural network and deep learning framework, and appropriate parameters are set. The model is trained through the training set, and the validation set is used regularly during the training to evaluate the model. The training strategy is organized according to the evaluation structure. When the performance of the model on the validation set meets expectations, the test set is used to conduct a final test on the model to evaluate the generalization ability of the model. The designed interactive page is convenient for inputting user design requirements to retrieve relevant ethnic cultural elements from the database, and use the trained deep learning model to generate design sketches. Designers further view and edit and adjust through the interactive page. After the designer further deepens the sketch, the ethnic costume design drawing is delivered to the user.
[0060] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.
Claims
1. A method for designing ethnic clothing based on deep learning, characterized in that: The following steps are involved: S1. Use Internet crawlers to collect images, patterns, colors and other data of various ethnic costumes on the network platform, including photos of traditional costumes, illustrations of costumes in historical documents, etc., cooperate with photos taken in ethnic areas during field investigations, and cooperate with ethnic cultural research institutions and scholars to ensure the accuracy and professionalism of the data obtained; S2. Organize a professional design team and ethnic culture experts to annotate the collected data, including ethnic categories, clothing styles, pattern types, color matching, craft characteristics, cultural implications, etc., and specify detailed annotation specifications and standards to ensure the consistency and accuracy of the annotation; S3. Perform image denoising, image enhancement, image segmentation and other pre-processing operations on the annotated image data through image recognition technology to improve image quality, clean and organize the acquired text data, remove irrelevant and redundant information, and convert it into a structured data format to associate and integrate with the image data; S4. Select TensorFlow as the deep learning framework, combine the microbial agent neural network with multiple convolutional layers, pooling layers and fully connected layers, and continuously adjust the parameters and optimize the algorithm to build a deep learning model. Combine the residual connection and global average pooling layer to improve the accuracy and generalization ability of the model. After daily training, the model is optimized by the algorithm to facilitate learning the characteristics and style of national costumes; S5. By establishing an interactive interface, users can associate clothing types, ethnic style preferences, color matching requirements and pattern element selection according to design requirements. After the user describes the design details, after text parsing and processing, it is converted into information coding and data signals that can be understood by the model, and a preliminary sketch is generated through comprehensive analysis and processing by the deep learning model; S6. After obtaining the design sketches produced in S5, the designer evaluates the clothing styles, further optimizes some details, and forms a complete ethnic clothing design plan in conjunction with the annotated ethnic cultural information.
2. The method for designing ethnic clothing based on deep learning according to claim 1, characterized in that: The data on images, patterns, colors, etc. of ethnic costumes obtained in S1 are the main source of information for establishing an ethnic culture database, and are obtained from various museums and ethnic culture research institutions at home and abroad, field photography and research in ethnic concentrated areas, and the use of web crawler technology to screen and select appropriate ethnic costume data.
3. The method for designing ethnic clothing based on deep learning according to claim 1, characterized in that: In the process of labeling the acquired data, the S2 simultaneously assists in the establishment of the national culture database, and after labeling all acquired data, the labeled data is classified and sorted.
4. The method for designing ethnic clothing based on deep learning according to claim 2, characterized in that: The ethnic culture database is classified in such a way that clothing data belonging to the same ethnic group are grouped together to form a sub-data set based on ethnicity. In each ethnic sub-data set, the data is further subdivided according to clothing styles. A separate pattern database is established in the ethnic culture database to organize all marked patterns according to type.
5. The method for designing ethnic clothing based on deep learning according to claim 4 is characterized in that: The colors of the clothing data in the ethnic culture database are classified according to the characteristics of the color, such as hue, brightness, saturation, etc., and the clothing craftsmanship is classified according to different production processes, and the data in the ethnic culture database is assisted by data annotation by a professional team.
6. The method for designing ethnic clothing based on deep learning according to claim 5, characterized in that: The professional team includes experts and fashion designers who have in-depth research on the history, cultural connotations, traditional customs, etc. of ethnic costumes and are familiar with different ethnic cultures, and they classify and label the ethnic types and styles of clothing.
7. The method for designing ethnic clothing based on deep learning according to claim 6, characterized in that: The professional team uses ethnological research results and official classifications to determine the typical characteristics of each ethnic group's clothing as the basis for judgment. For ethnic category labeling, they develop clear ethnic division standards, clarify labeling categories and standards, and establish an audit mechanism to review labeling.
8. The method for designing ethnic clothing based on deep learning according to claim 1, characterized in that: The convolution layer in the deep learning algorithm model in S4 has multiple convolution kernels of different scales, the convolution layer adopts a depth-separable convolution structure, and the residual connection is combined to process the residual information between the convolution layers. The output layer in the deep learning algorithm model adopts a softmax activation function to output the probability distribution of different clothing style categories.
9. The method for designing ethnic clothing based on deep learning according to claim 1, characterized in that: The model training of the deep learning algorithm model relies on providing learning materials from 70% training sets, 20% validation sets and 10% test sets divided in the national culture database, and selects the cross entropy loss function to provide image prediction result training for clothing style classification.
10. The method for designing ethnic clothing based on deep learning according to claim 9, characterized in that: The cross entropy loss function uses the Adam adaptive learning rate optimization algorithm to balance the model training parameters during the training process, and adds performance indicators to monitor the number of training rounds on the observed validation set.