Intelligent embroidery pattern generation method and system based on machine learning model

Abstract

The invention discloses an intelligent embroidery pattern generation method and system based on a machine learning model, and the method comprises the steps: collecting embroidery pattern design demands in real time according to an e-commerce platform, social media and market trend data, and obtaining a real-time updated data set; according to the data set, multi-modal features of the embroidery patterns are extracted, pattern feature identifiers with cultural features are generated, and fused pattern features are obtained; according to the fused pattern features, generating an embroidery pattern meeting the design requirements of the minority, and obtaining a preliminary embroidery pattern; and according to the preliminary embroidery pattern, compressing the small sample generation model into a lightweight model suitable for small and medium-sized enterprises or mobile terminal equipment so as to obtain an embroidery pattern generated in real time. According to the embodiment of the invention, market demands can be responded in real time, small embroidery patterns with cultural characteristics and high quality are generated, and the practical applicability of the generated model on small and medium-sized enterprises and mobile terminal equipment is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning, and particularly relates to a method and system for intelligent embroidery pattern generation based on a machine learning model. Background Art

[0002] Driven by the trends of globalization and personalized consumption, embroidery, as a traditional craft art form, is gradually integrating into the modern design field and becoming an important tool for brand communication and cultural expression. However, traditional embroidery design methods often rely on the rich experience and creativity of manual artists and are difficult to meet the rapidly changing market demands and the trend of personalized customization. Therefore, developing an intelligent embroidery pattern generation method that can effectively improve the flexibility and efficiency of embroidery design has become an important topic in the current industry. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for intelligent embroidery pattern generation based on a machine learning model to solve the deficiencies in the prior art, be able to respond to market demands in real time, generate niche embroidery patterns with cultural characteristics and high quality, and improve the practical applicability of the generation model on small and medium-sized enterprises and mobile devices.

[0004] An embodiment of the present application provides a method for intelligent embroidery pattern generation based on a machine learning model, and the method includes: According to e-commerce platforms, social media, and market trend data, collect embroidery pattern design requirements in real time, and through streaming data processing, combined with an incremental learning algorithm, dynamically update the training data set to obtain a real-time updated data set; According to the real-time updated data set, extract multi-modal features of the embroidery pattern, where the multi-modal features include color, texture, and cultural elements. Among them, the feature extraction uses a multi-modal fusion model based on a graph convolutional network, combined with a style transfer algorithm, to generate a pattern feature identifier with cultural characteristics and obtain the fused pattern features; According to the fused pattern features, use a few-shot generation model based on meta-learning to generate embroidery patterns that meet niche design requirements. Among them, the few-shot generation model optimizes the generation quality under few-shot inputs through attention technology and adaptive weight adjustment technology to obtain preliminary embroidery patterns; According to the preliminary embroidery patterns, use a lightweight model compression technology based on knowledge distillation to compress the few-shot generation model into a lightweight model suitable for small and medium-sized enterprises or mobile devices. Among them, the compression is achieved through hierarchical distillation and quantization technology to ensure the balance between generation speed and memory occupancy to obtain real-time generated embroidery patterns.

[0005] Optionally, the method further includes: Based on the feedback data of the user on the generated embroidery patterns, a model optimization algorithm based on reinforcement learning is adopted to dynamically adjust the parameters of the small-sample generation model; wherein, the optimization is achieved through a real-time feedback loop and adaptive learning rate adjustment, continuously improving the market adaptability of the generated embroidery patterns, and obtaining an optimized small-sample generation model.

[0006] Optionally, according to the e-commerce platform, social media, and market trend data, the design requirements of embroidery patterns are collected in real time. Through stream data processing and combined with the incremental learning algorithm, the training data set is dynamically updated to obtain a real-time updated data set, including: According to the e-commerce platform, social media, and market trend data, a data collection framework based on stream data processing is adopted to collect multi-source data on the design requirements of embroidery patterns in real time. Through lightweight data caching technology, the real-time and continuity of data collection are ensured. For the collected multi-source data, a noise filtering method based on the adaptive filtering algorithm is adopted. Combining the data format and content features, noise data and redundant information are removed. Through dynamic threshold adjustment technology, a preliminary cleaned data set is generated. For the preliminary cleaned data set, a data set update method based on the incremental learning algorithm is adopted. Combining historical training data and real-time data streams, the training data set is dynamically updated. Through data distribution balancing technology, the diversity and timeliness of the data set are ensured, and a preliminary real-time updated data set is generated. For the preliminary real-time updated data set, a verification method based on data consistency verification is adopted. Combining the data format and content features, the integrity and consistency of the data set are verified. Through feedback correction technology, a final real-time updated data set is generated.

[0007] Optionally, according to the real-time updated data set, multi-modal features of embroidery patterns are extracted. The multi-modal features include color, texture, and cultural elements. Among them, the feature extraction adopts a multi-modal fusion model based on the graph convolutional network, combined with the style transfer algorithm, to generate a pattern feature identifier with cultural characteristics, and obtain the fused pattern features, including: For the real-time updated data set, a multi-modal data preprocessing method based on deep learning is adopted to extract the preliminary features of color, texture, and cultural elements respectively. Through the adaptive data cleaning algorithm, a preliminary multi-modal feature representation is generated. For the preliminary multi-modal feature representation, a feature extraction method based on the graph convolutional network is adopted. The color, texture, and cultural elements are abstracted as nodes in the graph structure, and the association relationship between nodes is abstracted as edges. Through dynamic graph structure learning technology, a preliminary graph structure feature representation is generated. For the preliminary graphic structure feature representation, a feature fusion method based on the style transfer algorithm is adopted, combined with cultural element features, to generate a pattern feature identifier with cultural characteristics. Through cross-modal attention technology, the correlation relationship between different modal features is captured to generate a preliminary fused feature representation; For the preliminary fused feature representation, an optimization method based on error feedback technology is adopted, combined with real-time data streams and historical data distributions, to dynamically adjust the feature weights. Through regularization constraints, overfitting is prevented to generate the final fused pattern features.

[0008] Optionally, based on the fused pattern features, a few-shot generation model based on meta-learning is adopted to generate embroidery patterns that meet the needs of niche designs. Among them, the few-shot generation model optimizes the generation quality under few-shot inputs through attention technology and adaptive weight adjustment technology to obtain preliminary embroidery patterns, including: For the fused pattern features, a few-shot data preprocessing method based on meta-learning is adopted, combined with niche design requirements, to generate a preliminary few-shot dataset. Through data augmentation technology, the diversity of the few-shot dataset is expanded to generate a preliminary augmented dataset; For the preliminary augmented dataset, a few-shot generation model based on meta-learning is adopted, combined with attention technology and adaptive weight adjustment technology, to dynamically adjust the model parameters. Through multiple rounds of iterative optimization, a preliminary training model is generated; For the preliminary training model, a pattern generation method based on a generative adversarial network is adopted, combined with niche design requirements, to generate preliminary embroidery patterns. Through adaptive weight adjustment technology, the quality of the generated patterns is optimized to generate preliminary optimized patterns; For the preliminary optimized patterns, a verification method based on user feedback is adopted, combined with niche design requirements and real-time performance monitoring, to dynamically adjust the pattern design. Through feedback correction technology, the final preliminary embroidery patterns are generated.

[0009] Optionally, based on the preliminary embroidery patterns, a lightweight model compression technology based on knowledge distillation is adopted to compress the few-shot generation model into a lightweight model suitable for small and medium-sized enterprises or mobile devices. Among them, the compression is achieved through hierarchical distillation and quantization technology to ensure the balance between generation speed and memory occupancy to obtain real-time generated embroidery patterns, including: For the preliminary embroidery patterns, a hierarchical distillation method based on knowledge distillation is adopted, combined with the structure of the few-shot generation model, to layer-by-layer compress the model parameters. Through dynamic weight allocation technology, the accuracy and stability of the distillation process are ensured to generate a preliminary distillation model; For the preliminary distillation model, a model compression method based on quantization technology is adopted, combined with the model parameters and generation speed requirements, to dynamically adjust the quantization accuracy. Through adaptive quantization threshold adjustment technology, a preliminary compressed model is generated; For the preliminary compression model, a verification method based on simulation is adopted. Combining the generation speed and memory occupancy requirements, the performance of the model is verified. Through the feedback correction technology, the model parameters are dynamically adjusted to generate a preliminary optimized model. For the preliminary optimized model, a pattern generation method based on real-time generation technology is adopted. Combining the requirements of small and medium-sized enterprises or mobile devices, real-time embroidery patterns are generated. Through real-time monitoring technology, the stability and efficiency of the generation process are ensured to generate the final real-time generated embroidery patterns.

[0010] Another embodiment of the present application provides an intelligent embroidery pattern generation system based on a machine learning model. The system includes: An acquisition module, configured to collect real-time embroidery pattern design requirements according to e-commerce platforms, social media, and market trend data. Through stream data processing and combining incremental learning algorithms, the training data set is dynamically updated to obtain a real-time updated data set. An extraction module, configured to extract multi-modal features of embroidery patterns according to the real-time updated data set. The multi-modal features include colors, textures, and cultural elements. Among them, the feature extraction adopts a multi-modal fusion model based on a graph convolutional network and combines a style transfer algorithm to generate a pattern feature identifier with cultural characteristics to obtain the fused pattern features. A generation module, configured to generate embroidery patterns adapted to niche design requirements according to the fused pattern features by using a few-shot generation model based on meta-learning. Among them, the few-shot generation model optimizes the generation quality under few-shot inputs through attention technology and adaptive weight adjustment technology to obtain preliminary embroidery patterns. A compression module, configured to compress the few-shot generation model into a lightweight model suitable for small and medium-sized enterprises or mobile devices by using a lightweight model compression technology based on knowledge distillation according to the preliminary embroidery patterns. Among them, the compression is achieved through hierarchical distillation and quantization technologies to ensure the balance between generation speed and memory occupancy to obtain real-time generated embroidery patterns.

[0011] Another embodiment of the present application provides a storage medium in which a computer program is stored. Among them, the computer program is set to execute the method described in any one of the above when running.

[0012] Another embodiment of the present application provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is set to run the computer program to execute the method described in any one of the above.

[0013] Compared with the prior art, a method for generating intelligent embroidery patterns based on a machine learning model provided by the present invention collects the design requirements of embroidery patterns in real time according to e-commerce platforms, social media, and market trend data, and obtains a real-time updated data set; according to the data set, multi-modal features of the embroidery patterns are extracted to generate pattern feature identifiers with cultural characteristics, and the fused pattern features are obtained; according to the fused pattern features, embroidery patterns adapted to niche design requirements are generated to obtain preliminary embroidery patterns; according to the preliminary embroidery patterns, a small-sample generation model is compressed into a lightweight model applicable to small and medium-sized enterprises or mobile devices, and real-time generated embroidery patterns can be obtained, so as to be able to respond to market demands in real time, generate niche embroidery patterns with cultural characteristics and high quality, and improve the practical applicability of the generation model on small and medium-sized enterprises and mobile devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 FIG. is a block diagram of the hardware structure of a computer terminal for a method for generating intelligent embroidery patterns based on a machine learning model provided by an embodiment of the present invention; Figure 2 FIG. is a schematic flowchart of a method for generating intelligent embroidery patterns based on a machine learning model provided by an embodiment of the present invention; Figure 3 FIG. is a schematic structural diagram of a system for generating intelligent embroidery patterns based on a machine learning model provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0016] An embodiment of the present invention first provides a method for generating intelligent embroidery patterns based on a machine learning model. This method can be applied to electronic devices, such as computer terminals, specifically ordinary computers, etc.

[0017] The following takes running on a computer terminal as an example to explain it in detail. Figure 1 FIG. is a block diagram of the hardware structure of a computer terminal for a method for generating intelligent embroidery patterns based on a machine learning model provided by an embodiment of the present invention. As Figure 1 shown, the computer device includes a processor, a memory, and a network interface connected through a system bus. Among them, the memory may include a non-volatile storage medium and an internal memory.

[0018] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions. When the program instructions are executed, the processor can execute any method for generating intelligent embroidery patterns based on a machine learning model.

[0019] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0020] The internal memory provides an environment for the operation of a computer program in a non-volatile storage medium. When the computer program is executed by the processor, the processor can be enabled to execute any intelligent embroidery pattern generation method based on a machine learning model.

[0021] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art can understand that Figure 1 the structure shown in

[0022] is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0023] See Figure 2 , an embodiment of the present invention provides an intelligent embroidery pattern generation method based on a machine learning model, which may include the following steps: S201, according to e-commerce platform, social media and market trend data, collect embroidery pattern design requirements in real time, through streaming data processing, combined with incremental learning algorithm, dynamically update the training data set, and obtain a real-time updated data set; The core of this step is to use modern data collection technology to monitor and analyze the demand for embroidery patterns in the market in real time. Through the comprehensive data of e-commerce platforms and social media, the changes in consumer preferences and market trends can be obtained in a timely manner. This process includes extracting information from multiple data sources (such as user reviews, purchase records, interactions on social media, etc.) and cleaning and standardizing it through streaming data processing technology to ensure the real-time and accuracy of the data. Combined with the incremental learning algorithm, a new data set can be gradually constructed on the basis of the existing training data, avoiding repeated training and improving the learning efficiency of the model.

[0024] The significance of this process lies in providing a dynamic and real-time updated data support platform for embroidery pattern design. By collecting and analyzing market demands in a timely manner, designers can better grasp the popular trends and consumer preferences, which helps to create embroidery patterns that better meet the market demands. At the same time, the incremental learning method ensures the continuous improvement of the model, making the generated pattern designs more accurate and more competitive in the market.

[0025] Specifically, according to the e-commerce platforms, social media, and market trend data, a data collection framework based on stream data processing can be adopted to collect multi-source data on the demand for embroidery pattern design in real time. Through lightweight data caching technology, the real-time and continuity of data collection can be ensured; In this step, a framework based on stream data processing is used to monitor and collect data from e-commerce platforms, social media, and market trend analysis in real time. These data can include multiple information sources such as user reviews, purchase behaviors, and chat discussions. The implementation of lightweight data caching technology will ensure that during the data collection process, even during peak periods, the speed and stability of data collection can be maintained, thus avoiding data loss or delay caused by system overload. The implementation of this step guarantees the real-time nature of the data, enabling the design team to instantly obtain the changes in consumers' demands for embroidery patterns, thereby adjusting the design strategy more flexibly and responding to market dynamics and user preferences in a timely manner. This real-time feedback mechanism will enhance the success rate of product design and ensure that the final product can attract the target user group.

[0026] First, a data collection framework based on stream data processing needs to be constructed to monitor and obtain multi-source data from e-commerce platforms, social media, and market trends in real time. This process can be achieved by using stream processing tools such as Apache Kafka. During the data collection process, the system will call the API interfaces of major e-commerce platforms to capture relevant data on embroidery patterns in real time, including sales volume, customer reviews, and forum discussions. At the same time, the tags and posts on social media regarding embroidery patterns will also be automatically collected. To ensure the availability of the data, the team can use lightweight data caching technology (such as Redis) to temporarily store this data, enabling the system to still maintain a fast response under high traffic conditions. In addition, a visualization panel for real-time data monitoring will be developed to facilitate the team to promptly understand market dynamics and changes in user demands.

[0027] After data collection is completed, based on the incremental learning algorithm, the machine learning model will update the training dataset in real time. The advantage of incremental learning is that it can update the entire model without retraining when new data is received, but only needs to train on the new data. This process not only speeds up the model update speed but also ensures the timeliness of the dataset. For example, when a specific embroidery pattern becomes popular on social media, the incremental learning model can quickly capture this new information and automatically adjust its parameters to improve the prediction accuracy of the pattern. Therefore, the ability to obtain the latest market demand responses at any time enables the team to quickly adjust the design plan and product line.

[0028] During the dynamic update process, the diversity and richness of data are crucial. To ensure this, the system will automatically evaluate the quality and effectiveness of data sources and prioritize collecting data from platforms with greater influence. For example, if the sales volume of a certain embroidery pattern surges on a well-known e-commerce platform while the feedback on other platforms is average, the system will adjust the weights to ensure that the dataset can fully reflect the real demands of the main market. In addition, to avoid data redundancy and noise, the team should also regularly review the data collection strategy to ensure that the information collected is relevant and useful. Ultimately, this step will build a real-time updated training dataset that can reflect market dynamics, providing a solid foundation for subsequent feature extraction and model training.

[0029] Another implementation method: In this step, the design team needs to build an efficient data collection architecture to quickly collect the design requirements of embroidery patterns from multiple channels. First, a streaming data processing framework (such as Apache Kafka or Apache Flink) can be selected. These frameworks can process data streams from e-commerce platforms (such as Amazon, Taobao) and social media (such as Weibo, Instagram). By setting up corresponding API interfaces, the team can obtain data such as user comments, purchase behaviors, and popularity trends of embroidery patterns in real time. For example, when a certain embroidery pattern suddenly receives attention on social media, the system can automatically capture the relevant discussions and incorporate this data into subsequent analysis.

[0030] Second, to solve the data processing problem in high-concurrency situations, lightweight data caching technology will be introduced. Taking Redis as an example, this in-memory database system can quickly access and store data, ensuring high response speed even during peak periods (such as promotional activities). In practice, the system can design a caching strategy to temporarily store the latest comments and purchase records in memory for quick access by subsequent analysis modules, avoiding delays caused by directly extracting data from the database. If it is found that the sales volume of a certain embroidery pattern surges on the e-commerce platform, the data in the cache will enable the design team to obtain immediate feedback so as to quickly adjust the design strategy.

[0031] Finally, the success of real-time data collection not only depends on the implementation of technology but also requires continuous monitoring and optimization. The design team can set monitoring metrics for data collection, such as data latency time, success rate, etc., to ensure that the system can operate smoothly under any circumstances. In addition, the team should regularly evaluate the reliability of data sources to ensure the accuracy and timeliness of the collected data. For example, by regularly reviewing the sales statistics of e-commerce platforms and combining user feedback and trend analysis on social media, the team can form a complete data loop to provide real-time market basis for the design of embroidery patterns.

[0032] For the multi-source data collected, adopt a noise filtering method based on the adaptive filtering algorithm, combine data format and content features, remove noise data and redundant information, and generate a preliminary cleaned data set through dynamic threshold adjustment technology; A large amount of the collected data usually contains a lot of noise and redundant information, which may interfere with subsequent analysis and model training. The noise filtering method based on the adaptive filtering algorithm can clean the data by setting a dynamic threshold to automatically identify and remove irrelevant noise data and duplicate information. This method depends on data content and format features to ensure the quality and accuracy of the cleaned data. This step not only improves the quality of the data set but also provides a solid foundation for subsequent feature extraction and model training. The cleaned data set can effectively improve the training efficiency and effect of the model, reduce the learning errors caused by inaccurate data, and thus make the finally generated embroidery patterns more market-adaptive.

[0033] Focus on data cleaning to ensure high-quality data for subsequent analysis and model training. First, through the adaptive filtering algorithm, the system will filter the noise of the original data. For example, on social media, users may post a large amount of irrelevant or low-quality information, and the system can screen the data by setting specific filtering criteria, such as sentiment scores, comment lengths, etc. This method can effectively remove those worthless information and improve the overall quality of the data. In this process, the team can also use sentiment analysis technology to identify positive and negative comments on the embroidery patterns to further optimize the construction of the data set.

[0034] As data cleaning progresses, the team needs to combine data format and content features and use dynamic threshold adjustment technology to ensure that the cleaned data set can accurately reflect changes in market demand. For example, before a certain festival, users' interest in embroidery patterns of a specific theme may increase sharply, and the team can automatically increase the filtering sensitivity for relevant data to ensure that these new data are preferentially retained. The resulting preliminary cleaned data set will be more accurate, and the information contained will be more capable of guiding design directions and production strategies.

[0035] Finally, the preliminary dataset processed by the adaptive filtering technology will have high reliability and can effectively support subsequent feature extraction and model construction. In practical applications, the team can further improve the effectiveness of the dataset by regularly auditing the cleaning process and adjusting the data filtering strategy according to market feedback. In addition, the system can be designed to automatically learn and optimize during the data flow process, making the future data cleaning process more efficient and intelligent. Therefore, this process not only improves the quality of the data but also provides a solid foundation for subsequent data analysis.

[0036] Another implementation method: In the second step, the design team will implement a noise filtering process to improve the quality of the dataset. First, the collected data will be analyzed and cleaned through an adaptive filtering algorithm (such as applying Kalman filtering). The team can first determine which data features are crucial for embroidery design, for example, the sentiment score of comments, the purchase frequency of users, etc. The system will use these features as a basis and identify and remove irrelevant noise by setting a series of rules. For example, if a comment is long but does not mention the quality or style of the embroidery pattern, the system will mark it as deletable content.

[0037] Next, the team will apply dynamic threshold adjustment technology in combination with data format and content features. Using real-time data flow, the system can automatically analyze and update the filtering criteria to adapt to the data characteristics at different times. For example, during a holiday, user preferences may change, resulting in a significant increase in certain types of comments. At this time, the dynamic threshold will adjust the sensitivity of noise recognition to ensure that the cleaning process can adapt to data changes in real time, retain valid information, and remove redundant details.

[0038] By implementing this process, the finally generated preliminary cleaned dataset will lay a good foundation for subsequent analysis. This dataset only contains highly relevant information, making the subsequent feature extraction and model training processes more efficient. In practical applications, the cleaned dataset will help the design team more accurately identify the popular trends of embroidery patterns. For example, if the cleaned dataset shows that a certain color of embroidery pattern has received wide acclaim during a specific festival, the design team can quickly adjust its product line and launch corresponding new embroidery patterns.

[0039] For the preliminary cleaned dataset, adopt a dataset update method based on the incremental learning algorithm, combine historical training data and real-time data flow, dynamically update the training dataset, and ensure the diversity and timeliness of the dataset through data distribution balancing technology to generate a preliminary real-time updated dataset; In this step, the preliminarily cleaned data is combined with the historical training data through an incremental learning algorithm to achieve dynamic dataset update. Such a technique can quickly adapt to new data streams without having to train the model from scratch. The data distribution balancing technique ensures that during the update process, model training will not be biased due to an excessive or insufficient amount of data in a certain category, thus enhancing the comprehensiveness and accuracy of the model. By flexibly updating the training dataset, the model can better reflect the current market demands and user preferences, ensuring that the designed embroidery patterns keep up with the times. This approach greatly improves the model's response speed and adaptability, enabling it to promptly capture the demands of popular trends and consumer interests, thereby enhancing the market performance of the final product.

[0040] The dynamically updated dataset of the preliminarily cleaned data will be utilized by means of an incremental learning algorithm. The focus of this process lies in combining the new real-time data stream with the historical training data, and through continuous learning of the model, maintaining the vitality and accuracy of the dataset. By establishing a dynamic update mechanism, the system can transparently integrate the new data into the existing training dataset when it receives new data and perform necessary feature adjustments. In this way, the model can quickly adapt to market changes without complex retraining steps, further reducing the consumption of computing resources.

[0041] To ensure the diversity and timeliness of the dataset, the team will adopt the data distribution balancing technique. Specifically, the system will analyze the distribution of various categories in the dataset and actively adjust the structure of the dataset according to the preset balancing criteria. For example, if the sample size of a certain embroidery pattern is significantly excessive, the system will randomly select some samples to ensure that the data volumes of different design styles are balanced, avoiding the model's bias towards the dominant category. In this way, the model can more comprehensively reflect the market demands, enhancing the diversity and adaptability of the generated embroidery patterns.

[0042] Ultimately, through dynamic update and distribution balancing, the generated real-time updated dataset will better reflect the actual demand trends of the market. The design team can carry out more precise embroidery pattern design based on this dataset. For example, when it is detected that the discussion degree of a certain type of design has increased significantly on different social media platforms, the team can respond promptly and conduct the development and promotion of relevant designs. This not only improves the market competitiveness of the product but also enhances the team's adaptability in the dynamic market environment.

[0043] Another implementation approach: In this step, the design team will focus on dynamically updating the preliminarily cleaned dataset using incremental learning algorithms. Incremental learning enables the model to process new data without having to retrain all the data, which is particularly suitable for handling streaming data. In specific implementation, the team can first merge the real-time data stream with the historical training data to form a dynamically updated training dataset. To this end, the system needs to periodically check the validity and representativeness of the new data. For example, by setting thresholds to determine which new data should be included in the training set and which should be ignored.

[0044] Then, to ensure the diversity and timeliness of the dataset, the team will implement data distribution balancing techniques. Specifically, the system can automatically evaluate the distribution of data across various categories and adjust the dataset through sampling or undersampling techniques to avoid model bias caused by an excessive amount of data of a certain type of embroidery pattern. For example, if the number of training samples of a certain embroidery pattern is significantly higher than that of other styles, the team can choose to increase the samples of this type in the new data or balance the overall structure of the dataset by extracting some samples. In this way, even if the market demand for embroidery patterns changes, the model can remain adaptable.

[0045] Finally, as the real-time data stream continues to be added, the design team needs to continuously monitor and evaluate the performance of the updated training dataset to ensure that it can effectively reflect the current market situation. In practice, the team can set up an automated feedback mechanism to regularly check model performance metrics such as accuracy and recall. If it is found that the model performs poorly in a specific area, the system will automatically initiate the incremental learning mechanism to adjust the dataset again and update the model to ensure that the design team can always keep abreast of market trends and timely adjust its embroidery pattern development strategy.

[0046] For the preliminary real-time updated dataset, a verification method based on data consistency check is adopted. Combining data format and content features, verify the integrity and consistency of the dataset, and generate the final real-time updated dataset through feedback correction technology.

[0047] After the data is updated, it is crucial to ensure the consistency and integrity of the dataset. This step verifies the preliminarily updated dataset through data consistency check methods, examines the consistency of the data in terms of format and content features, and promptly discovers and corrects any possible problems. Through feedback correction technology, the system can self-learn and optimize the dataset to ensure its quality and reliability. Through this step, the finally generated dataset will be more complete and consistent. A high-quality dataset not only helps improve the accuracy of model training but also enhances the quality of the generated embroidery patterns. This lays a good foundation for subsequent feature extraction and model training, ensuring that the generated embroidery patterns can better meet market demands.

[0048] Data consistency verification will be performed on the preliminarily real-time updated dataset to ensure data integrity and consistency. First of all, setting a series of data consistency criteria is crucial, which includes requirements for field integrity, such as the user ID, review text, rating, etc. must be available, and checking whether the data types and formats of these fields meet expectations. For example, the rating field should only contain numerical data within a certain range; while the review text field needs to ensure no missing values and be of text type. By setting these criteria, any data that does not meet the requirements will be automatically marked for correction.

[0049] Subsequently, once inconsistencies or errors are found in the dataset, the system will use feedback correction technology to guide them back to the cleaning module. For example, if the timestamp of a certain review shows a future date, the system will automatically mark it and delete it or request manual review after recognition. This feedback loop not only ensures data quality but also provides a basis for continuous improvement in data cleaning and verification. Through this automated feedback correction mechanism, the team can solve data problems more quickly, maintaining the integrity and consistency of the dataset continuously.

[0050] Finally, after the above data consistency verification and feedback correction, the generated final real-time updated dataset will have the characteristics of high quality and can effectively support the subsequent deep learning model training. The design team can use this dataset to conduct accurate market analysis and design decisions. For example, when the final dataset shows that a certain embroidery pattern has good sales on multiple e-commerce platforms, the team can quickly promote the product based on this information. In addition, the team can continuously optimize the model and design strategy according to the changes in the dataset to improve market adaptability and user satisfaction, thus gaining an advantage in the competition.

[0051] S202, Extract multi-modal features of the embroidery pattern according to the real-time updated dataset. The multi-modal features include color, texture, and cultural elements. Among them, the feature extraction uses a multi-modal fusion model based on a graph convolutional network, combined with a style transfer algorithm, to generate a pattern feature identifier with cultural characteristics and obtain the fused pattern features; This step aims to extract important features that affect the embroidery pattern design from the real-time updated dataset. Through the graph convolutional network (GCN), the model can effectively process graph data and extract the associations and features between different modalities (such as color, texture, and cultural elements). In addition, combined with the style transfer algorithm, the system can fuse design elements from different cultural backgrounds, thus generating embroidery pattern features with unique cultural values. This method not only considers the diversity and complexity of the patterns but also provides a rich feature basis for subsequent pattern generation.

[0052] Extracting pattern features with cultural characteristics plays an important role in embroidery pattern design. First of all, rich multi-modal features enable the model to comprehensively understand market trends and user preferences, ensuring that the designed embroidery patterns meet the needs and aesthetics of consumers. Secondly, by integrating different cultural elements, the design team can create embroidery patterns with connotations and stories, thereby enhancing the added value of products and strengthening market competitiveness. In addition, the fused pattern features provide a solid foundation for the subsequent generation process, making the generated embroidery patterns not only novel and unique, but also able to arouse the emotional resonance of consumers, thus increasing the sales conversion rate.

[0053] Specifically, for a real-time updated dataset, a multi-modal data preprocessing method based on deep learning can be adopted to extract the preliminary features of color, texture, and cultural elements respectively, and generate a preliminary multi-modal feature representation through an adaptive data cleaning algorithm. In this step, the system preprocesses the real-time updated dataset through deep learning techniques, aiming to extract the basic features of embroidery patterns, mainly including color, texture, and cultural elements. By using convolutional neural networks (CNNs) and other deep learning algorithms, the system can automatically identify and extract these important features. The introduction of the adaptive data cleaning algorithm can effectively filter out useless information, ensuring that the extracted features are clearer and more accurate. The significance of this process lies in laying a good foundation for subsequent feature fusion and generation. Through the extracted multi-modal preliminary features, the model can better understand the relationships between various features, making the subsequent feature fusion more effective. At the same time, accurate feature extraction also provides the necessary support for generating high-quality embroidery patterns, ensuring that the designed patterns can meet the needs of consumers.

[0054] When implementing this step, the design team first needs to ensure that the input dataset is properly formatted to suit the input requirements of deep learning algorithms. The team aggregates data collected from various channels (such as product pictures on e-commerce platforms, user-uploaded content on social media, etc.). In this step, data collection should not only pay attention to the uniformity of format, but also eliminate low-quality and redundant information to avoid affecting the accuracy of subsequent feature extraction.

[0055] Next, the team will apply a deep convolutional neural network (CNN), which can effectively extract features such as color, texture, and cultural elements from images. For example, for a picture of an embroidery pattern, the model first identifies its color distribution and extracts the main colors and color tone information. Secondly, the texture features can analyze the fineness and complexity of the pattern through filters in reverse, while the cultural element features will focus on identifying traditional patterns, symbols, or texts therein. The key in this process is to use pre-trained models, such as VGG or ResNet, to accelerate feature extraction and improve its accuracy.

[0056] Finally, the team applied an adaptive data cleaning algorithm, combined with the initially extracted features, to generate a multi-modal feature representation. The adaptive data cleaning algorithm can automatically identify and process noisy data according to the types of features in the dataset. For example, if there are outliers in a certain color feature in the samples, the algorithm will automatically mark and remove these data points that do not conform to the rules. Through this series of operations, the team generated a preliminary multi-modal feature representation covering a variety of important features, providing a solid foundation for subsequent analysis and generation.

[0057] For the preliminary multi-modal feature representation, a feature extraction method based on graph convolutional network was adopted. The color, texture, and cultural elements were abstracted as nodes in the graph structure, and the association relationships between the nodes were abstracted as edges. Through the dynamic graph structure learning technology, a preliminary graph structure feature representation was generated; In this step, a graph convolutional network (GCN) was used to further process the initially extracted multi-modal features. The color, texture, and cultural elements were regarded as nodes in the graph, and the association relationships between them were abstracted as edges between the nodes. Through the graph convolutional network, the system can capture the interactions between these nodes and update and optimize the features through the dynamic graph structure learning technology.

[0058] The key role of this step is to enhance the model's ability to understand the relationships between complex features. By transforming the features into a graph structure, the model can more effectively capture the interdependent relationships between different modal features, helping to generate more delicate embroidery patterns. In addition, the dynamic learning ability of the graph structure improves the flexibility and adaptability of the model in dealing with complex designs, ensuring that the generated patterns can better reflect the actual design requirements.

[0059] In this process, the team first transformed the preliminary multi-modal feature representation into a graph structure for in-depth feature analysis using the graph convolutional network (GCN) method. Each type of feature, including color, texture, and cultural elements, would be represented as a node in the graph. For example, the color node might include red, blue, and green, etc., while the texture node could reflect the fineness or ruggedness of the embroidery. The team needed to clarify the relationships between the nodes and associate them in the form of edges. These edges could represent the degree of mutual influence between the nodes, such as the preference of a certain cultural element for a specific color.

[0060] Next, the team used a graph convolutional network to extract features from the graph structure. Through GCN, the model can learn the relationships between adjacent nodes and aggregate based on local node information. For example, the features of a certain texture node not only depend on its own features but are also affected by the surrounding color nodes and cultural element nodes. Through multiple graph convolution operations, the team can extract higher-level feature representations, forming deep-level information of the graph structure features. This way of feature abstraction enables the model to better understand the complex relationships between features and achieve a more artistic and culturally profound effect when generating embroidery patterns.

[0061] Finally, the team applied dynamic graph structure learning technology to further optimize the obtained features. This step allows the model to dynamically update the connections and weights between features as new data is added. For example, in the expression of cultural elements, if a certain traditional embroidery pattern suddenly becomes popular, the system will increase the weights of the related feature nodes, reflecting the new trend. Through such a dynamic learning process, the generated preliminary graph structure feature representation will be more time-sensitive and accurate, laying a foundation for subsequent style transfer and feature fusion.

[0062] For the preliminary graph structure feature representation, a feature fusion method based on a style transfer algorithm is adopted. Combining cultural element features, a pattern feature identifier with cultural characteristics is generated. Through cross-modal attention technology, the correlation relationships between different modal features are captured to generate a preliminary fused feature representation; In this step, the team used a style transfer algorithm to fuse the preliminary graph structure features, with the focus on combining color, texture, and cultural element features. Through the style transfer algorithm, the model can automatically transform and fuse the styles of different features, thus generating a pattern expression with a unique cultural charm. In addition, the introduction of cross-modal attention technology enables the model to focus on the important feature relationships between different modalities, forming a more refined fused feature representation.

[0063] This process plays a crucial role in enhancing the cultural depth and artistry of embroidery patterns. By effectively fusing different modal features, the team can create embroidery designs with cultural connotations, thereby enhancing the competitiveness and market attractiveness of products. In addition, the application of cross-modal attention technology also enables the model to pay more attention to user preferences and needs during the generation process, ensuring that the final pattern design is more targeted.

[0064] In this step, the team will process the preliminary graphical structure feature representation with a style transfer algorithm to achieve feature fusion. First, the team defines the target style and content image. The target style image can be a traditional embroidery work with a specific cultural background, while the content image is the pattern extracted from the preliminary graphical structure features. Through the style transfer algorithm, the model can apply the features in the target style to the content image, thereby generating a new pattern feature identifier that integrates cultural characteristics.

[0065] Next, the team uses cross-modal attention technology to enhance the interaction between features. Specifically, the team assigns attention weights to each feature (color, texture, and cultural elements), enabling the model to focus on the important relationships between different modalities. For example, when generating an embroidery themed on a certain traditional culture, the system will automatically increase the weights of the texture and color nodes related to it, while weakening the influence of irrelevant features. Such a feature selection process can ensure that the finally generated pattern not only has rich expressiveness but also is closer to cultural characteristics.

[0066] Finally, after being processed by the style transfer and attention mechanism, the generated pattern feature identifier with cultural characteristics will play an important role in the subsequent generation process. In this way, the team can create embroidery patterns that combine artistic value and market attractiveness. Such characteristics can not only promote design diversity but also enhance the competitiveness of products in the cultural market. In addition, embroidery patterns with cultural depth can trigger emotional resonance among users, thereby increasing consumers' willingness to purchase.

[0067] For the preliminary fused feature representation, an optimization method based on error feedback technology is adopted. Combining real-time data streams and historical data distributions, the feature weights are dynamically adjusted. Through regularization constraints, overfitting is prevented, and the final fused pattern features are generated.

[0068] In the last step, the team adopts an optimization method based on error feedback technology to dynamically adjust the preliminary fused feature representation. By combining real-time data streams and historical data, the model can update the feature weights in a timely manner to better reflect market demands. At the same time, using regularization constraints, the model can effectively prevent overfitting and ensure that the finally generated features have good generalization ability.

[0069] The core significance of this optimization process lies in improving the quality and market adaptability of the finally generated patterns. Through the dynamically adjusted feature weights, the model can more sensitively capture changes in the consumer market, enabling the generated embroidery patterns to quickly adapt to different design requirements. In addition, the application of regularization technology also improves the stability of the system, ensuring that after multiple iterations, the feature representation still maintains strong effectiveness and accuracy.

[0070] In this step, the team first conducts error feedback analysis on the preliminary fused feature representation to identify potential deficiencies in the model during the generation process. The team will establish a real-time feedback mechanism that uses user feedback data and sales data to monitor the market performance of the generated patterns. By comparing with the historical data distribution, the model can identify which features perform poorly in actual applications and then make corresponding adjustments.

[0071] Next, by dynamically adjusting the feature weights, the team can optimize the pattern features in real time. For example, if user feedback shows that a certain feature combination has a good response in the market, the model will correspondingly increase the weight of this part of the features while reducing the weights of other unpopular features. The core of this process lies in combining real-time data streams, enabling the model to be more flexible when processing new features. Through such a feedback mechanism, the team ensures that the finally generated pattern features can closely follow market dynamics and meet consumer needs.

[0072] Finally, to prevent overfitting, the team introduces regularization constraints. Regularization techniques can help control the complexity of the model, reduce overfitting to a specific dataset, and thus improve the generalization ability of the model. For example, the team may use L2 regularization to limit the magnitude of the feature weights to ensure that the model does not lose balance due to the excessive enhancement of certain features during feature adjustment. Through this series of optimizations and adjustments, the finally generated fused pattern features will have stronger market adaptability and usability, providing a solid foundation for the generation of embroidery patterns.

[0073] S203. According to the fused pattern features, use a few-shot generation model based on meta-learning to generate embroidery patterns that meet the needs of niche design. Among them, the few-shot generation model optimizes the generation quality under few-shot inputs through attention technology and adaptive weight adjustment technology to obtain preliminary embroidery patterns; The core of this step is to use the fused features and, through the method of meta-learning, effectively handle few-shot scenarios. Especially in niche design requirements, the model needs to efficiently generate patterns. Under the meta-learning framework, the system can learn how to extract valuable information from very limited data, thereby improving the performance of the generation model. The significance of this process lies in promoting design innovation and diversity. Especially in the generation of embroidery patterns in the niche market, it can effectively meet the personalized needs of specific users. By focusing on the performance of the few-shot generation model and applying the attention mechanism and adaptive weight adjustment, the system not only ensures the quality of pattern generation but also improves the flexibility and adaptability of the model. This will significantly enhance the market competitiveness of the product and enable the design team to quickly respond to market changes and create more embroidery works that meet consumer needs.

[0074] Specifically, for the fused pattern features, a few-shot data preprocessing method based on meta-learning can be adopted. Combining with the niche design requirements, a preliminary few-shot dataset can be generated. Through data augmentation techniques, the diversity of the few-shot dataset can be expanded to generate a preliminary augmented dataset. In this step, first, few-shot data preprocessing is performed on the fused pattern features to ensure that the generated embroidery patterns can meet the specific niche market requirements. This process uses the method of meta-learning to extract knowledge from previous data and applies this knowledge to generate a preliminary few-shot dataset. Next, the diversity of this few-shot dataset is further expanded through data augmentation techniques. For example, the original patterns can be transformed by methods such as rotation, scaling, cropping, and color transformation, so that the generated samples are not limited to the original design but also include diverse variants, thereby improving the generalization ability of the model.

[0075] Through this step, the generated few-shot dataset can more comprehensively represent the niche design requirements and enhance the adaptability of the model under specific design styles. Data augmentation not only increases the richness of training samples but also helps the model learn different design elements and combination methods. In this way, when facing actual design requirements, the model can quickly adapt and generate embroidery patterns that meet market preferences. Therefore, this process is of great significance in improving model performance and production efficiency.

[0076] In this process, first, in-depth analysis is performed on the fused pattern features to identify key elements such as colors, textures, and cultural symbols. Through market research and historical sales data, the trend of current niche design requirements is determined to form a preliminary design requirement framework. For example, if the system recognizes that a certain ethnic style of embroidery pattern is popular, this category can be set as the key direction for the subsequent few-shot dataset. Then, according to this requirement framework, potential samples are selected from the existing pattern library or designer creations to form a preliminary few-shot dataset.

[0077] After generating the preliminary dataset, the next task is to use data augmentation techniques to expand the diversity of this dataset. Multiple data augmentation methods can be adopted here, such as rotation, cropping, flipping, adjusting brightness and contrast, etc., to artificially increase the sample size and richness. For example, for a specific embroidery pattern, by rotating it at different angles (such as 90 degrees, 180 degrees, etc.) or displaying it in different backgrounds, the diversity of the samples can be effectively increased. Through these means, a dataset that may originally have only dozens of samples can be quickly expanded to hundreds or even thousands of data samples, providing a sufficient basis for subsequent model training.

[0078] Finally, after being processed by data augmentation techniques, an enhanced dataset with rich diversity is formed. This dataset not only includes variants of the original samples but also integrates the inspired patterns created by designers to ensure that the generated embroidery patterns can fully reflect the needs of the niche market. These enhanced data will provide more comprehensive support for subsequent model training, ensuring higher quality in the style and effect of the generated embroidery patterns.

[0079] For the preliminary enhanced dataset, a few-shot generation model based on meta-learning is adopted. Combining attention techniques and adaptive weight adjustment techniques, the model parameters are dynamically adjusted, and through multiple rounds of iterative optimization, a preliminary training model is generated. In this step, the researchers will use the meta-learning-based generation model to conduct in-depth research on the preliminary enhanced dataset. Through the attention mechanism, the model can focus on the important features in the dataset and concentrate on the key details that affect the design. At the same time, combined with the adaptive weight adjustment technique, the model can dynamically adjust the weights of each input feature during the training process to improve the quality and accuracy of the generated samples. Through multiple rounds of iterative optimization, the model continuously learns and improves, thus generating a preliminary training model that can effectively meet the needs of niche design. The core role of this step is to enhance the flexibility and adaptability of the generation model. Through the attention mechanism, the model can accurately capture the most representative features in the embroidery pattern and naturally integrate them into the generated pattern. The adjustment of adaptive weights ensures that the model can maintain good performance in various design requirements, and the finally generated embroidery patterns can better meet the needs of the niche market. This process significantly improves the diversity and uniqueness of the generated patterns, providing rich creative inspiration for designers.

[0080] In this step, based on the preliminary enhanced dataset, a few-shot generation model is constructed. First, the input layer of the model needs to be designed to receive multi-modal features, including color, texture, and cultural elements. To enable the model to better understand the needs of niche design, a meta-learning framework can be adopted when constructing the model, endowing it with the ability to quickly learn from a small number of examples. The key to meta-learning lies in the task assignment during the training process, enabling the model to learn to transfer between different design tasks. For example, in a design task with an ethnic style, how to use the existing embroidery samples to generate new patterns.

[0081] With the construction of the model, the next key lies in combining the attention mechanism to deeply analyze the input data. The attention mechanism allows the model to focus on the most important parts of the input features during the generation process. For example, under specific design requirements, the features of a certain color or pattern are particularly prominent. By comparing the feature weights between different samples, the model can adaptively adjust the generation strategy. For instance, if the user particularly likes a certain color combination, the model can preferentially retain these colors during generation, thereby generating a more personalized pattern.

[0082] During the model training process, through multiple rounds of iterative optimization, the generation effect is continuously improved. After each round, the model adjusts its parameters according to the feedback between the generated results and the real samples. For example, if an embroidery style generated fails to meet the standards expected by the market, the model will analyze these results and appropriately reduce the generation weight of this style in the next round. After multiple repeated iterations and adjustments, the generation model can finally generate embroidery patterns with highly personalized and aesthetic features under small-sample inputs, forming a preliminary training model.

[0083] For the preliminary training model, a pattern generation method based on a generative adversarial network is adopted, combined with niche design requirements, to generate preliminary embroidery patterns. Through the adaptive weight adjustment technology, the quality of the generated patterns is optimized to generate preliminary optimized patterns; In this step, a generative adversarial network (GAN) is introduced to generate preliminary embroidery patterns. In this model, two neural networks - the generator and the discriminator - are trained in an adversarial manner. The generator is dedicated to generating real embroidery patterns, while the discriminator is responsible for distinguishing the differences between the generated patterns and the real patterns. By combining niche design requirements, the generator can generate corresponding patterns according to specific requirements put forward by designers. At the same time, using the adaptive weight adjustment technology, the model can dynamically adjust the weights of relevant features during the generation process according to the generation effect, ensuring that the quality of the finally generated patterns meets the expected standards. The significance of this step is to improve the generation quality and diversification degree of embroidery patterns through the generative adversarial network. The design of GAN makes each generated pattern undergo a strict review, and only the patterns verified by the discriminator can be considered "high-quality". The combination of niche design requirements ensures that the generated patterns can better meet a specific market, increasing the personalization and attractiveness of the products. Finally, this process not only enriches the styles and forms of embroidery patterns but also provides a broader creative space for designers.

[0084] In the process of implementing pattern generation based on a Generative Adversarial Network (GAN), it is first necessary to build the basic structure of the GAN, including two parts: a generator and a discriminator. The task of the generator is to generate embroidery patterns based on the input small-sample data, while the discriminator is responsible for evaluating the quality of the generated patterns and determining the differences between them and real samples. In this process, the generator needs to make full use of the multi-modal features extracted from the previously trained model to ensure that the generated patterns can effectively meet the design requirements of niche markets.

[0085] Next, the generator starts to use the input data, using random noise as the initial input. After multiple layers of network calculations, it gradually generates preliminary embroidery patterns. To improve the quality of the generated patterns, combining the adaptive weight adjustment technique allows the generator to dynamically adjust the generation weights of each feature during the generation process according to the discriminator's feedback. For example, if the generator finds that a certain type of texture is not ideal in the discriminator's evaluation, it can reduce the generation weight of this texture while strengthening the weights of other more popular features, thus gradually optimizing the generated patterns.

[0086] In this process, repeated training and evaluation are essential. The generator continuously generates and updates patterns, while the discriminator outputs feedback on the generated patterns, and the generation effect is constantly in an improved state. Through this method, not only is the diversity of the generated patterns ensured, but also the balance between the artistic and commercial nature of the generated embroidery patterns is structurally guaranteed. Finally, after multiple rounds of feedback and optimization, the generator can produce a series of preliminary optimized patterns that fully represent niche needs and are also competitive in the market.

[0087] For the preliminary optimized patterns, a verification method based on user feedback is adopted. Combining niche design requirements and real-time performance monitoring, the pattern design is dynamically adjusted. Through the feedback correction technique, the final preliminary embroidery patterns are generated.

[0088] At this stage, the preliminary optimized embroidery patterns will be submitted to users for feedback. Based on the users' experiences and evaluations in actual use, the system will use the collected feedback data to verify the actual effects of the patterns. Combining niche design requirements and the performance data of real-time monitoring, the system can dynamically adjust the pattern design to ensure that the generated embroidery patterns better meet user expectations. This process usually uses the feedback correction technique. By analyzing and processing the data of user feedback, the pattern design is gradually optimized and corrected to form the final embroidery patterns. The key to this step is to integrate the real needs and feedback of users into the design process, thereby improving the practicality and market adaptability of the embroidery patterns. Through user feedback, designers can accurately grasp the interest points and popular trends of the target market, and then flexibly adjust in the design to improve the quality and aesthetics of the product. Finally, users not only get an embroidery pattern that meets their needs, but designers can also obtain more intuitive market feedback, forming a virtuous cycle.

[0089] In this step, a user feedback system is first established. After receiving the preliminary optimized pattern, users can evaluate and provide feedback through this system. The feedback content can include the aesthetics, practicality of the pattern, and the customer's recognition of cultural elements, etc. The collected feedback data will be integrated and analyzed for reference in subsequent designs. For example, if users indicate in the evaluation that the color combination of a certain pattern is unsatisfactory, the system will record this information and prioritize its handling.

[0090] Combining the niche design requirements and real-time performance monitoring, the system can dynamically adjust the preliminary optimized pattern. Through real-time monitoring tools, the system can instantly track the acceptance of different patterns by users. This means that once the system discovers that a certain style of pattern is significantly more popular among users than other styles, it can adjust the design strategy in the next round of generation and focus on improving this style. In this process, the algorithm can quickly respond to market changes to ensure that the design always aligns with user needs.

[0091] Finally, using feedback correction technology, designers will regularly review the user feedback data and fine-tune the pattern design. Designers can refine the elements of the pattern according to the feedback information to make it more in line with the needs of the niche market. For example, if users feedback that a certain embroidery style is not suitable enough in a specific occasion, designers can adjust the embroidery pattern to a more general or multi-functional design style based on this feedback. In this way, the finally generated embroidery pattern can not only better meet market demands but also improve user satisfaction, forming a good design closed-loop.

[0092] S204, according to the preliminary embroidery pattern, adopt the lightweight model compression technology based on knowledge distillation to compress the small-sample generation model into a lightweight model applicable to small and medium-sized enterprises or mobile devices. Among them, the compression is achieved through hierarchical distillation and quantization techniques to ensure the balance between the generation speed and memory occupancy, so as to obtain the real-time generated embroidery pattern.

[0093] According to the preliminary generated embroidery pattern, adopt the lightweight model compression technology based on knowledge distillation to compress the small-sample generation model into a lightweight model applicable to small and medium-sized enterprises or mobile devices. This process aims to simplify the original model with higher complexity and larger computational requirements into an efficient and easily deployable version to meet the resource limitations of small and medium-sized enterprises in design and production. At the same time, knowledge distillation technology is used to extract and retain the most valuable features in the model while ensuring the generation quality, reducing the computational complexity of the model.

[0094] The direct significance of this compression process lies in enhancing the practicality and operability of the model, enabling small and medium-sized enterprises or users on mobile devices to quickly and effectively generate embroidery patterns. This lightweight model can not only perform efficient pattern generation under limited resources but also shorten the waiting time for users, thus improving the user experience. In addition, by ensuring the balance between generation speed and memory occupancy, enterprises can reduce operating costs and achieve the ability to quickly respond to market changes.

[0095] Specifically, for the preliminary embroidery pattern, a hierarchical distillation method based on knowledge distillation can be adopted, combined with the structure of the small-sample generation model, to compress the model parameters layer by layer. Through dynamic weight allocation technology, the accuracy and stability of the distillation process are ensured, and a preliminary distilled model is generated. In this step, the hierarchical distillation method based on knowledge distillation mainly trains a simple student model to mimic the behavior of a complex teacher model. By gradually passing the parameters of the teacher model to the student model, the gradual compression of the model is achieved, ensuring that key information and features are retained as much as possible during the compression process. At the same time, hierarchical distillation can avoid information loss caused by directly compressing the entire model at once, laying a foundation for subsequent model optimization.

[0096] The successful implementation of this step can reduce the storage requirements and running time of the model, making the generation process more efficient. By reducing the model parameters of each layer, the student model can complete the task of pattern generation at a lower computational cost, thus better adapting to the resource limitations of small and medium-sized enterprises and mobile devices. At the same time, this method can ensure that the generated embroidery patterns still maintain high quality and rich details, so as to have a certain competitiveness in the market.

[0097] In this step, first, a structural analysis of the preliminary embroidery pattern generation model needs to be carried out to identify different levels of the model and their importance to the output results. By combining hierarchical distillation technology, the knowledge of each layer in the teacher model can be extracted layer by layer and passed on to the student model. For example, in a small-sample generation model, there may be multiple convolutional layers and fully connected layers. We can preferentially select convolutional layers for distillation because convolutional layers play a key role in feature extraction and pattern generation. By implementing "layer-by-layer distillation", it can be ensured that the output of the student model at each layer matches well with the teacher model.

[0098] When performing the distillation process, the dynamic weight allocation technique plays an important role. The core of this technique lies in dynamically adjusting the weights of different layers in the loss function according to their importance. For example, if the output of a certain layer has a greater impact on the generated pattern, a higher weight can be assigned to this layer, thereby promoting the student model to learn more feature information. In this way, the distillation process will become more accurate and stable, ensuring that key features are not lost during the compression process.

[0099] Finally, by monitoring the performance of the student model, it is evaluated whether the generated embroidery pattern meets the expected standards. If the generated pattern has high quality and is similar to the output of the teacher model, it can be considered that the preliminary distillation model has been successfully generated. Further tests and optimizations can be carried out subsequently to ensure the operability and reliability of the model in actual applications, enabling the lightweight model to adapt to different application scenarios.

[0100] For the preliminary distillation model, a model compression method based on quantization technology is adopted. Combining the model parameters and the requirements of the generation speed, the quantization precision is dynamically adjusted. Through the adaptive quantization threshold adjustment technology, a preliminary compressed model is generated; In this process, quantization technology is used to convert the floating-point numerical model parameters into integers with lower bit numbers to reduce the memory occupancy and computational complexity of the model. According to the actual generation speed requirements of the model, an appropriate quantization precision is selected, and the quantization parameters of each layer are dynamically adjusted to ensure that the performance and generation quality of the model are retained as much as possible during the processing. Through the application of quantization technology, the model can maintain a certain generation ability after compression, while significantly reducing the storage requirements, which is particularly important for small and medium-sized enterprises and mobile device users with limited resources. This compression method ensures real-time generation ability, helps enterprises improve the response speed and flexibility during product development and market promotion, and at the same time maintains the quality of the user experience.

[0101] At the beginning of the compression process, a comprehensive evaluation of the preliminary distillation model is required to determine the quantization requirements of each layer's parameters and the requirements of the generation speed. Quantization technology usually involves converting floating-point numbers into integer representations with lower bit numbers, thereby reducing the memory occupancy and computational complexity of the model. For this purpose, a dynamic adaptive quantization strategy can be used to analyze the weights and activation values of each layer to determine the appropriate quantization precision. For example, for relatively critical convolutional layers, a higher quantization precision may be required, while for layers with less impact, a lower quantization precision can be adopted.

[0102] In specific operations, first, multiple tests can be conducted to record the performance data of the model at different quantization precisions. Based on these data, an adaptive quantization threshold adjustment technique can be further developed to adjust the quantization precision of each layer in real time according to the performance of the model during operation. This means that when a certain layer detects slow generation speed, the system can automatically reduce the quantization precision of that layer to improve the generation speed. Conversely, when the generation quality of a certain layer is affected, its quantization precision can be increased to ensure the quality of the generated results.

[0103] After the initial compressed model is generated, the model still needs to be verified and evaluated to determine whether it meets the requirements of real-time generation. This can be comprehensively evaluated by setting performance metrics (such as generation speed, memory occupancy, etc.) and ensuring that the model performs consistently on different devices. Through such dynamic adjustment and optimization, it is ensured that the finally generated embroidery pattern can operate efficiently on small and medium-sized enterprises or mobile devices.

[0104] For the initial compressed model, a verification method based on simulation is adopted. Combining the generation speed and memory occupancy requirements, the performance of the model is verified. Through feedback correction technology, the model parameters are dynamically adjusted to generate a preliminary optimized model; In this step, simulation is used to verify whether the compressed model meets the requirements of generation speed and memory occupancy. By setting a series of simulation scenarios, the generation performance of the model under different conditions is evaluated, such as response time and resource consumption, so as to ensure the efficient operation of the model in actual applications. Through simulation verification, the risk of running in a real environment can be significantly reduced, ensuring that the compressed model is not only theoretically feasible but also can achieve the expected effect in actual applications. This process can help the development team timely discover potential problems of the model under specific conditions and make further adjustments and optimizations, thereby improving the stability and reliability of the model before it is put on the market.

[0105] In this step, a simulation environment needs to be built for the initial compressed model to test its performance under different conditions. First, a series of test cases are designed, which include requests for generating embroidery patterns with different complexities. By simulating various user inputs, the system can establish a relatively real usage scenario to test the generation speed and memory occupancy of the model. For example, different pattern requirements can be set from simple to complex, and the time required and resources occupied by the model during generation can be measured.

[0106] After obtaining these performance data, analyze the feedback results of the generation speed and memory usage. Usually, it may be found that the memory usage of some layers exceeds the preset limit, or the generation speed fails to meet the expectations. At this time, the feedback correction technology will come into play, allowing developers to make necessary adjustments to the model parameters according to the detected performance bottlenecks. By comparing the test data with the user requirements, dynamically adjust the model parameters, such as resetting the quantization threshold, optimizing the calculation method of specific layers, etc., to achieve a balance between the generation speed and memory usage.

[0107] Finally, through repeated simulation verification and parameter adjustment, an optimized model is finally generated. This optimized model can not only meet the user requirements in terms of performance, but also has the ability to adapt to market changes. Ensure that on small and medium-sized enterprises and mobile devices, the model can stably and efficiently generate high-quality embroidery patterns to meet various challenges that may be encountered in actual application scenarios.

[0108] For the preliminary optimized model, adopt a pattern generation method based on real-time generation technology, combine the needs of small and medium-sized enterprises or mobile devices, generate real-time embroidery patterns, and ensure the stability and efficiency of the generation process through real-time monitoring technology to generate the final real-time generated embroidery patterns.

[0109] In this step, the optimized model will be applied to the task of real-time generating embroidery patterns. Based on real-time generation technology, the model can quickly respond to the input needs of users and generate personalized embroidery patterns. This method must fully consider the operating conditions of small and medium-sized enterprises and mobile devices to ensure a smooth and efficient generation process. Real-time monitoring technology plays an important role in this process, being able to track the generation progress in real time and promptly discover and handle possible performance bottlenecks. The implementation of this step enables users to enjoy a fast and instant embroidery pattern generation experience, which is especially suitable for small and medium-sized enterprises and individual users with rapidly changing market demands. Through real-time generation, enterprises can better respond to market feedback, provide personalized products, thereby enhancing customer satisfaction and market competitiveness. At the same time, monitor the stability and efficiency of the generation process to ensure that the production line of the tool can operate continuously and stably, thereby reducing operation risks and costs.

[0110] When implementing real-time generation technology, first build an efficient architecture suitable for small and medium-sized enterprises and mobile devices. This architecture needs to support streaming data processing and a model with strong scene adaptation ability, which can quickly generate embroidery patterns according to user requirements. For example, the data access, processing, and output modules can be decoupled through a microservices architecture, thereby enhancing the flexibility and adaptability of the system and enabling each module to operate efficiently and independently.

[0111] Next, real-time monitoring technology will be used to track various metrics during the generation process, such as generation time, resource consumption, etc., to ensure a balance between generation speed and quality. Through real-time monitoring, the system can quickly detect potential performance issues, such as a sudden drop in generation speed or excessive memory usage, etc., and react immediately, such as adjusting resource allocation or optimizing the priority order of model calls, to ensure the smoothness of the overall process.

[0112] Finally, the system needs to provide a dynamic user feedback mechanism to ensure that users can obtain real-time generation status and results. After the user submits an embroidery pattern request, the system can update the generation progress in real time, allowing the user to understand the situation at each stage and display the results in a timely manner after generation is completed. This interactive experience not only gives users a sense of participation but also continuously optimizes the generation strategy through real-time feedback to meet market demands and personalized design requirements, ensuring that the finally generated embroidery patterns reach an ideal state in terms of both quality and speed.

[0113] Specifically, in practical applications, based on the feedback data of users on the generated embroidery patterns, a model optimization algorithm based on reinforcement learning can be adopted to dynamically adjust the parameters of the small-sample generation model; wherein, the optimization is achieved through a real-time feedback loop and adaptive learning rate adjustment, continuously improving the market adaptability of the generated embroidery patterns to obtain an optimized small-sample generation model.

[0114] In this method, based on the feedback data of users on the generated embroidery patterns, a model optimization algorithm based on reinforcement learning is adopted to dynamically adjust the parameters of the small-sample generation model. Specifically, the system will collect evaluation data of users on each embroidery pattern, such as user likes, comments, or purchase behaviors, and convert these feedbacks into reward signals. By introducing the policy optimization framework of reinforcement learning, the model learns how to adjust its parameters according to these feedbacks to generate embroidery patterns that better meet market demands. The key to this process lies in the real-time feedback loop. With the help of adaptive learning rate adjustment technology, the system can continuously optimize the model while generating embroidery patterns, enabling it to adapt to the changing needs of users more quickly.

[0115] This optimization process is of great significance. First of all, through the integration of real-time feedback, the market adaptability of the generated patterns is improved, ensuring that the design not only reflects the current needs of users but also can respond promptly to changes in market trends. For example, if users give enthusiastic feedback on a certain style of pattern but a lower evaluation on another style of pattern, the model can quickly adjust the generation strategy, increasing the learning intensity of the popular style features, thereby improving the market acceptance of subsequent generated patterns. In this way, enterprises can continuously launch products that meet consumer preferences, increasing sales opportunities and brand loyalty.

[0116] Based on the feedback data from users on the generated embroidery patterns, a model optimization algorithm based on reinforcement learning is adopted to dynamically adjust the parameters of the small-sample generation model. This process first involves establishing a feedback collection mechanism that allows users to evaluate each generated embroidery pattern on the platform. Users can express their opinions by scoring, selecting like or dislike options, or even writing comments. These feedback messages will be collected in real time and stored in the database, forming a rich user feedback dataset. The system analyzes the users' feedback, regarding positive feedback as reward signals and negative feedback as punishment signals. By inputting these feedback signals into the reinforcement learning model, the model gradually learns under what circumstances the generated embroidery patterns are more popular and can be adjusted in real time according to changes in user preferences.

[0117] To achieve the optimal effect of dynamic adjustment, this method introduces a real-time feedback loop and an adaptive learning rate adjustment technique in the optimization process. Specifically, the core of model optimization lies in continuously updating the model's parameters through iterative interaction with user feedback. For example, when an embroidery pattern receives high praise from users, the system increases the weight of this design feature to strengthen the model's learning of similar styles. At the same time, if a certain feature continuously receives negative evaluations, the system quickly reduces its weight to reduce the influence of this element in future generations. This dynamic adjustment not only helps the model quickly adapt to users' needs but also enables it to continuously improve to ensure that the generated embroidery patterns have higher market adaptability and user satisfaction.

[0118] Finally, the adjustment of the adaptive learning rate makes the learning process more flexible and effective. Whenever the model receives new user feedback, the system evaluates the changes in the feedback and adjusts the model's learning rate based on this. For example, if a large amount of user feedback is collected in a short period, indicating that the market demand changes rapidly, the system may increase the learning rate to speed up the model's response to user preferences; conversely, if the feedback changes less, the system will reduce the learning rate to ensure the stability of the model during updates. This method not only improves the learning efficiency of the model but also ensures that the generated embroidery patterns can better meet users' expectations, ultimately forming a precise and fast-responsive generation system, further enhancing the enterprise's competitiveness in the market.

[0119] It can be seen that according to the e-commerce platform, social media and market trend data, the design requirements of embroidery patterns are collected in real time to obtain a real-time updated data set; according to the data set, the multi-modal features of the embroidery patterns are extracted, and a pattern feature identifier with cultural characteristics is generated to obtain the fused pattern features; according to the fused pattern features, embroidery patterns that meet the niche design requirements are generated to obtain preliminary embroidery patterns; according to the preliminary embroidery patterns, the small sample generation model is compressed into a lightweight model suitable for small and medium-sized enterprises or mobile devices, and real-time generated embroidery patterns can be obtained, so as to be able to respond to market demands in real time, generate niche embroidery patterns with cultural characteristics and high quality, and improve the practical applicability of the generation model on small and medium-sized enterprises and mobile devices.

[0120] Another embodiment of the present invention provides an intelligent embroidery pattern generation system based on a machine learning model. Refer to Figure 3 , the system may include: A collection module 301, configured to collect the design requirements of embroidery patterns in real time according to the e-commerce platform, social media and market trend data, and through streaming data processing, combined with an incremental learning algorithm, dynamically update the training data set to obtain a real-time updated data set; An extraction module 302, configured to extract the multi-modal features of the embroidery patterns according to the real-time updated data set. The multi-modal features include color, texture and cultural elements. Among them, the feature extraction uses a multi-modal fusion model based on a graph convolutional network, combined with a style transfer algorithm, to generate a pattern feature identifier with cultural characteristics, and obtain the fused pattern features; A generation module 303, configured to generate embroidery patterns that meet the niche design requirements according to the fused pattern features by using a small sample generation model based on meta-learning. Among them, the small sample generation model optimizes the generation quality under small sample input through an attention technology and an adaptive weight adjustment technology to obtain preliminary embroidery patterns; A compression module 304, configured to compress the small sample generation model into a lightweight model suitable for small and medium-sized enterprises or mobile devices according to the preliminary embroidery patterns by using a lightweight model compression technology based on knowledge distillation. Among them, the compression is realized through hierarchical distillation and quantization technologies to ensure the balance between the generation speed and the memory occupation, so as to obtain real-time generated embroidery patterns.

[0121] It can be seen that according to the data of e-commerce platforms, social media, and market trends, the design requirements of embroidery patterns are collected in real time to obtain a real-time updated dataset; according to the dataset, the multi-modal features of the embroidery patterns are extracted, and a pattern feature identifier with cultural characteristics is generated to obtain the fused pattern features; according to the fused pattern features, embroidery patterns that meet the niche design requirements are generated to obtain preliminary embroidery patterns; according to the preliminary embroidery patterns, the small-sample generation model is compressed into a lightweight model suitable for small and medium-sized enterprises or mobile devices, and real-time generated embroidery patterns can be obtained, so as to be able to respond to market demands in real time, generate niche embroidery patterns with cultural characteristics and high quality, and improve the practical applicability of the generation model on small and medium-sized enterprises and mobile devices.

[0122] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, and the computer program is configured to execute the steps in any one of the above method embodiments when running.

[0123] Specifically, in this embodiment, the above storage medium may be configured to store a computer program for executing the following steps: S201, according to the data of e-commerce platforms, social media, and market trends, collect the design requirements of embroidery patterns in real time, and through streaming data processing, combined with the incremental learning algorithm, dynamically update the training dataset to obtain a real-time updated dataset; S202, according to the real-time updated dataset, extract the multi-modal features of the embroidery patterns, where the multi-modal features include color, texture, and cultural elements. Among them, the feature extraction uses a multi-modal fusion model based on a graph convolutional network, combined with a style transfer algorithm, to generate a pattern feature identifier with cultural characteristics to obtain the fused pattern features; S203, according to the fused pattern features, use a small-sample generation model based on meta-learning to generate embroidery patterns that meet the niche design requirements. Among them, the small-sample generation model optimizes the generation quality under small-sample input through attention technology and adaptive weight adjustment technology to obtain preliminary embroidery patterns; S204, according to the preliminary embroidery patterns, use a lightweight model compression technology based on knowledge distillation to compress the small-sample generation model into a lightweight model suitable for small and medium-sized enterprises or mobile devices. Among them, the compression is achieved through hierarchical distillation and quantization technology to ensure the balance between generation speed and memory occupancy to obtain real-time generated embroidery patterns.

[0124] It can be seen that, according to the e-commerce platform, social media, and market trend data, the design requirements of embroidery patterns are collected in real time, and a real-time updated dataset is obtained; according to the dataset, the multi-modal features of the embroidery patterns are extracted, a pattern feature identifier with cultural characteristics is generated, and the fused pattern features are obtained; according to the fused pattern features, embroidery patterns adapted to the niche design requirements are generated to obtain preliminary embroidery patterns; according to the preliminary embroidery patterns, the small-sample generation model is compressed into a lightweight model applicable to small and medium-sized enterprises or mobile devices, and real-time generated embroidery patterns can be obtained, so as to be able to respond to market demands in real time, generate niche embroidery patterns with cultural characteristics and high quality, and improve the practical applicability of the generation model on small and medium-sized enterprises and mobile devices.

[0125] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0126] Specifically, the above electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the above processor, and the input / output device is connected to the above processor.

[0127] Specifically, in this embodiment, the above processor may be configured to execute the following steps through a computer program: S201, according to the e-commerce platform, social media, and market trend data, collect the design requirements of embroidery patterns in real time, and through streaming data processing, combined with the incremental learning algorithm, dynamically update the training dataset to obtain a real-time updated dataset; S202, according to the real-time updated dataset, extract the multi-modal features of the embroidery patterns, the multi-modal features include color, texture, and cultural elements, wherein the feature extraction uses a multi-modal fusion model based on a graph convolutional network, combined with a style transfer algorithm, to generate a pattern feature identifier with cultural characteristics, and obtain the fused pattern features; S203, according to the fused pattern features, use a small-sample generation model based on meta-learning to generate embroidery patterns adapted to niche design requirements, wherein the small-sample generation model optimizes the generation quality under small-sample input through attention technology and adaptive weight adjustment technology to obtain preliminary embroidery patterns; S204, according to the preliminary embroidery patterns, use a lightweight model compression technology based on knowledge distillation to compress the small-sample generation model into a lightweight model applicable to small and medium-sized enterprises or mobile devices, wherein the compression is achieved through hierarchical distillation and quantization technologies to ensure the balance between generation speed and memory occupancy, so as to obtain real-time generated embroidery patterns.

[0128] It can be seen that according to the e-commerce platform, social media, and market trend data, the design requirements of embroidery patterns are collected in real time to obtain a real-time updated dataset; according to the dataset, the multi-modal features of embroidery patterns are extracted to generate a pattern feature identifier with cultural characteristics, and the fused pattern features are obtained; according to the fused pattern features, embroidery patterns adapted to niche design requirements are generated to obtain preliminary embroidery patterns; according to the preliminary embroidery patterns, the small-sample generation model is compressed into a lightweight model suitable for small and medium-sized enterprises or mobile devices, and real-time generated embroidery patterns can be obtained, so as to be able to respond to market demands in real time, generate niche embroidery patterns with cultural characteristics and high quality, and improve the practical applicability of the generation model on small and medium-sized enterprises and mobile devices.

[0129] The structure, features, and function effects of the present invention have been described in detail based on the embodiments shown in the drawings. The above is only the preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the drawings. Any changes made according to the concept of the present invention, or equivalent embodiments modified into equivalent changes, still within the spirit covered by the specification and drawings, shall be within the protection scope of the present invention.

Claims

1. An intelligent embroidery pattern generation method based on a machine learning model, characterized in that, The method includes: According to e-commerce platforms, social media, and market trend data, real-time collect the design requirements of embroidery patterns. Through stream data processing and combined with incremental learning algorithms, dynamically update the training data set to obtain a real-time updated data set; According to the real-time updated data set, extract the multi-modal features of embroidery patterns. The multi-modal features include color, texture, and cultural elements. Among them, feature extraction uses a multi-modal fusion model based on graph convolutional networks and combines style transfer algorithms to generate pattern feature identifiers with cultural characteristics to obtain the fused pattern features; According to the fused pattern features, use a few-shot generation model based on meta-learning to generate embroidery patterns that meet niche design requirements. Among them, the few-shot generation model optimizes the generation quality under few-shot inputs through attention technology and adaptive weight adjustment technology to obtain preliminary embroidery patterns; According to the preliminary embroidery patterns, use a lightweight model compression technology based on knowledge distillation to compress the few-shot generation model into a lightweight model suitable for small and medium-sized enterprises or mobile devices. Among them, the compression is achieved through hierarchical distillation and quantization technologies to ensure the balance between generation speed and memory occupancy to obtain real-time generated embroidery patterns.

2. The method according to claim 1, wherein The method further includes: According to the feedback data of users on the generated embroidery patterns, use a model optimization algorithm based on reinforcement learning to dynamically adjust the parameters of the few-shot generation model; among them, the optimization is achieved through real-time feedback loops and adaptive learning rate adjustment to continuously improve the market adaptability of the generated embroidery patterns to obtain an optimized few-shot generation model.

3. The method according to claim 2, wherein The real-time collection of embroidery pattern design requirements according to e-commerce platforms, social media, and market trend data, through stream data processing, combined with incremental learning algorithms, and dynamically updating the training data set to obtain a real-time updated data set includes: According to e-commerce platforms, social media, and market trend data, use a data collection framework based on stream data processing to real-time collect multi-source data on embroidery pattern design requirements. Through lightweight data caching technology, ensure the real-time and continuity of data collection; For the collected multi-source data, use a noise filtering method based on adaptive filtering algorithms, combined with data formats and content features, to remove noise data and redundant information. Through dynamic threshold adjustment technology, generate a preliminary cleaned data set; For the preliminary cleaned data set, use a data set update method based on incremental learning algorithms, combined with historical training data and real-time data streams, to dynamically update the training data set. Through data distribution balancing technology, ensure the diversity and timeliness of the data set to generate a preliminary real-time updated data set; For the preliminary real-time updated data set, use a verification method based on data consistency verification, combined with data formats and content features, to verify the integrity and consistency of the data set. Through feedback correction technology, generate the final real-time updated data set.

4. The method according to claim 3, wherein Extract the multi-modal features of the embroidery pattern according to the real-time updated data set. The multi-modal features include color, texture, and cultural elements. Among them, feature extraction uses a multi-modal fusion model based on a graph convolutional network, combined with a style transfer algorithm, to generate a pattern feature identifier with cultural characteristics, and obtain the fused pattern features, including: For the real-time updated data set, use a multi-modal data preprocessing method based on deep learning to extract the preliminary features of color, texture, and cultural elements respectively, and generate a preliminary multi-modal feature representation through an adaptive data cleaning algorithm; For the preliminary multi-modal feature representation, use a feature extraction method based on a graph convolutional network to abstract color, texture, and cultural elements as nodes in the graph structure, and abstract the association relationship between nodes as edges. Through dynamic graph structure learning technology, generate a preliminary graph structure feature representation; For the preliminary graph structure feature representation, use a feature fusion method based on a style transfer algorithm, combined with cultural element features, to generate a pattern feature identifier with cultural characteristics. Through cross-modal attention technology, capture the association relationship between different modal features, and generate a preliminary fused feature representation; For the preliminary fused feature representation, use an optimization method based on an error feedback technology, combined with the real-time data stream and historical data distribution, dynamically adjust the feature weights, and prevent overfitting through regularization constraints to generate the final fused pattern features.

5. The method according to claim 4, wherein According to the fused pattern features, use a few-shot generation model based on meta-learning to generate embroidery patterns that meet the needs of niche designs. Among them, the few-shot generation model optimizes the generation quality under few-shot inputs through attention technology and adaptive weight adjustment technology, and obtains preliminary embroidery patterns, including: For the fused pattern features, use a few-shot data preprocessing method based on meta-learning, combined with niche design requirements, to generate a preliminary few-shot data set, and expand the diversity of the few-shot data set through data augmentation technology to generate a preliminary augmented data set; For the preliminary augmented data set, use a few-shot generation model based on meta-learning, combined with attention technology and adaptive weight adjustment technology, dynamically adjust the model parameters, and generate a preliminary training model through multiple rounds of iterative optimization; For the preliminary training model, use a pattern generation method based on a generative adversarial network, combined with niche design requirements, to generate preliminary embroidery patterns, and optimize the quality of the generated patterns through adaptive weight adjustment technology to generate preliminary optimized patterns; For the preliminary optimized patterns, use a verification method based on user feedback, combined with niche design requirements and real-time performance monitoring, dynamically adjust the pattern design, and generate the final preliminary embroidery patterns through feedback correction technology.

6. The method according to claim 5, wherein According to the preliminary embroidery patterns, use a lightweight model compression technology based on knowledge distillation to compress the few-shot generation model into a lightweight model suitable for small and medium-sized enterprises or mobile devices. Among them, the compression is achieved through hierarchical distillation and quantization technology to ensure the balance between generation speed and memory occupancy, so as to obtain real-time generated embroidery patterns, including: For the preliminary embroidery pattern, a hierarchical distillation method based on knowledge distillation is adopted, combined with the structure of the few-shot generation model. The model parameters are compressed layer by layer, and through the dynamic weight allocation technology, the accuracy and stability of the distillation process are ensured to generate a preliminary distillation model; For the preliminary distillation model, a model compression method based on quantization technology is adopted, combined with the model parameters and the requirements of the generation speed. The quantization accuracy is dynamically adjusted, and through the adaptive quantization threshold adjustment technology, a preliminary compression model is generated; For the preliminary compression model, a verification method based on simulation is adopted, combined with the generation speed and the memory occupancy requirements to verify the performance of the model. Through the feedback correction technology, the model parameters are dynamically adjusted to generate a preliminary optimized model; For the preliminary optimized model, a pattern generation method based on real-time generation technology is adopted, combined with the requirements of small and medium-sized enterprises or mobile devices to generate real-time embroidery patterns. Through the real-time monitoring technology, the stability and efficiency of the generation process are ensured to generate the final real-time generated embroidery pattern.

7. An intelligent embroidery pattern generation system based on a machine learning model, characterized in that, The system includes: An acquisition module, configured to collect the design requirements of embroidery patterns in real time according to e-commerce platforms, social media, and market trend data. Through stream data processing, combined with the incremental learning algorithm, the training data set is dynamically updated to obtain a real-time updated data set; An extraction module, configured to extract the multi-modal features of the embroidery pattern according to the real-time updated data set. The multi-modal features include color, texture, and cultural elements. Among them, the feature extraction adopts a multi-modal fusion model based on a graph convolutional network, combined with a style transfer algorithm to generate a pattern feature identifier with cultural characteristics to obtain the fused pattern features; A generation module, configured to generate an embroidery pattern that meets the niche design requirements according to the fused pattern features by using a few-shot generation model based on meta-learning. Among them, the few-shot generation model optimizes the generation quality under the few-shot input through the attention technology and the adaptive weight adjustment technology to obtain a preliminary embroidery pattern; A compression module, configured to compress the few-shot generation model into a lightweight model applicable to small and medium-sized enterprises or mobile devices according to the preliminary embroidery pattern by using a lightweight model compression technology based on knowledge distillation. Among them, the compression is realized through hierarchical distillation and quantization technology to ensure the balance between the generation speed and the memory occupancy to obtain a real-time generated embroidery pattern.

8. The system according to claim 7, characterized in that, The system further includes: A feedback module, configured to dynamically adjust the parameters of the few-shot generation model by using a model optimization algorithm based on reinforcement learning according to the feedback data of the user on the generated embroidery pattern. Among them, the optimization is realized through a real-time feedback loop and an adaptive learning rate adjustment to continuously improve the market adaptability of the generated embroidery pattern to obtain an optimized few-shot generation model.

9. A storage medium, characterized in that, A computer program is stored in the storage medium, wherein the computer program is set to execute the method according to any one of claims 1-6 when running.

10. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is set to run the computer program to execute the method according to any one of claims 1-6.

Images