Personalized costume design scheme generation method based on artificial intelligence

Abstract

The invention relates to the technical field of intelligent clothing design, in particular to an artificial intelligence-based personalized clothing design scheme generation method, which comprises the following steps of S1, modeling and fusing multi-dimensional user portrait data; s2, generating a style evolution design engine; and S3, interactively feeding back the three-dimensional fitting in real time. According to the method, the accurate user portrait is constructed through multi-dimensional user portrait data modeling and fusion integration of information from different data sources, the environment context is acquired in real time in combination with a knowledge graph and a spatio-temporal data acquisition technology, personalized data support is provided for subsequent costume design, and the user experience is improved. The emotion analysis and natural language processing technology is applied to analyze the aesthetic tendency and style preference of a user, costume design meeting the user requirement is generated based on a deep learning model, it is ensured that each design meets the fashion trend and the user preference through real-time evaluation and optimization of a design sketch, and the user experience can be improved through personalized recommendation and diversified design. And the innovativeness and uniqueness of the design scheme are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent clothing design, and in particular to a method for generating personalized clothing design solutions based on artificial intelligence. Background Art

[0002] Personalized clothing design is a core area of ​​the fashion industry's digital transformation, involving complex requirements such as body shape adaptation, style matching, scenario-based dressing, and dynamic interactive experiences. Existing technologies often rely on user base measurement collection, style questionnaires, or historical purchase data mining, combined with parametric template libraries to generate preliminary design solutions. Some systems incorporate two-dimensional image rendering and virtual fitting technologies to enhance visualization. With the expansion of metaverse consumption scenarios, the popularization of flexible sensing technology, and the miniaturization of real-time environmental perception devices, personalized design requires a deep integration of multi-dimensional data streams such as physiological characteristics, behavioral habits, emotional preferences, and temporal and spatial context, while achieving high-fidelity and low-latency responses in virtual twins, dynamic simulation, and cross-terminal interaction. In particular, when addressing instant outfit recommendations, functional adaptation of sportswear, and cross-cultural aesthetic integration, the system must possess multimodal data fusion, physically accurate simulation, and real-time generation and optimization capabilities to meet the full-link design requirements from static display to dynamic interaction.

[0003] Existing technologies rely on structured data input for traditional user profiling and lack effective integration of heterogeneous data from multiple sources, such as social media images, voice feedback, and body language. Knowledge graph construction focuses on static attribute associations, making it difficult to dynamically capture how user aesthetic preferences evolve in response to environmental factors, emotions, and fashion trends. Two-dimensional design generation systems are limited by the principle of planar projection and cannot accurately map the impact of three-dimensional human body curvature on clothing tailoring. Early 3D modeling technologies relied on standard body shape databases and were poorly adaptable to specific body shapes and dynamic postures, resulting in fabric drape distortion during virtual try-ons. Existing rendering engines often use simplified material ball models, which cannot accurately simulate the anisotropic reflective properties of complex fabrics such as silk and leather under mixed lighting. Motion physics simulations often simplify fabric dynamics using a mass-spring system, making it difficult to reproduce the microscopic details of wrinkle formation and stress distribution. Regarding cross-platform compatibility, mobile devices and VR devices often use reduced-order models, resulting in a fragmented try-on experience. Furthermore, the design optimization process lacks a closed-loop correction mechanism driven by real-time user feedback, leading to excessive cognitive load when comparing multiple solutions. Therefore, we propose an AI-based method for generating personalized clothing design solutions. Summary of the Invention

[0004] In view of the above-mentioned shortcomings of the prior art, the first purpose of the present invention is to provide a method for generating personalized clothing design solutions based on artificial intelligence to solve the problems in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for generating personalized clothing design solutions based on artificial intelligence, comprising the following steps:

[0007] S1. Multi-dimensional user portrait data modeling and integration;

[0008] S2, generative style evolution design engine;

[0009] S3, real-time interactive feedback 3D fitting;

[0010] S4, intelligent iterative optimization and verification;

[0011] S5. End-to-end intelligent interactive management of the supply chain.

[0012] The present invention is further configured as follows: in the step S1, multi-dimensional user portrait data modeling and fusion:

[0013] S1.1. Utilizing multi-source heterogeneous data fusion and integration technology, we unified the modeling of questionnaires, social media behavior logs, and wearable device physiological indicators through a graph database. We then constructed user attribute nodes based on a knowledge graph algorithm, accurately mapping multi-dimensional portraits such as height, body shape, style, and color preferences.

[0014] S1.2. Utilize spatiotemporal data collection and edge computing to obtain weather, scene lighting, and activity trajectories in real time at the user terminal. Dynamically model the environmental context through a time-series graph neural network and integrate it with the user profile node.

[0015] S1.3. Using the RoBERTa pre-trained large model combined with multimodal emotion recognition, we input the user's natural language description, expressions, and voice intonation into the Transformer emotion classifier to extract aesthetic tendencies and style appeals, generate emotion embedding vectors, and optimize the association with the user profile.

[0016] S1.4. Use structured light and TOF cameras to obtain high-precision depth maps, use point cloud reconstruction and shape repair algorithms to generate a 3D human mesh model, and perform parameter annotation and correction based on subdivision surfaces and ergonomic standards.

[0017] The present invention is further configured as follows: in the step S2, the generative style evolution design engine:

[0018] S2.1. Based on a convolutional neural network and a multi-head self-attention mechanism, we extract structured style features (silhouettes, patterns, decorative elements, etc.) from massive design galleries and fashion trend data, and dynamically update the style prototype library through a style clustering algorithm.

[0019] S2.2. We build a conditional GAN ​​(cGAN) combined with a variational autoencoder (VAE). This approach uses user profiles and environment vectors as conditional inputs, generates diverse design sketches through a cross-alignment loss and a multi-level discriminator, and evaluates the generated results in real-time using the FID and LPIPS metrics.

[0020] S2.3. Introducing wearable deformation simulation based on finite element analysis (FEA) into the generated garment sketch, automatically adjusting fabric shrinkage, stress distribution, and wrinkle effects through physical constraints and shape optimization iterative algorithms;

[0021] S2.4. Utilize collaborative filtering and deep matrix decomposition algorithms to integrate user sentiment embedding with popular color models (Pantone trend color library), output the optimal color combination and fabric material ratio, and generate realistic material maps through GAN, which can be directly used for virtual rendering or production samples.

[0022] The present invention is further configured as follows: in the step S3, real-time interactive feedback of three-dimensional fitting:

[0023] S3.1, combining PBR (Physically Based Rendering) and real-time ray tracing technology, allows 3D try-on in the browser based on WebGL or Unity engine, supporting fabric gravity, collision and muscle binding simulation, allowing users to intuitively evaluate the wearing effect under different lighting and movements;

[0024] S3.2. Using motion capture or mobile phone inertial navigation data, the virtual human skeleton and fabric physics engine are driven in real time to generate non-static fitting scenes based on the user's daily movements. Spatiotemporal smoothing filtering is also used to ensure smooth system response and natural simulation.

[0025] S3.3, supports multimodal interaction methods such as touch screen gestures, voice and gesture recognition. Users can fine-tune clothing details through voice commands or gestures. The system predicts and optimizes interactive behaviors based on a reinforcement learning strategy network;

[0026] S3.4. Using online A / B testing and the multi-armed bandit algorithm (Multi-Armed Bandit), we dynamically collect user clicks, dwell time, and secondary editing behaviors during the try-on process, and calculate the satisfaction score in real time through Bayesian optimization to guide the next round of solution iteration.

[0027] The present invention is further configured as follows: in the step S4, intelligent iterative optimization verification:

[0028] S4.1. Input user fitting, interaction, and satisfaction logs into the Transformer-XL time series neural network to perform behavior sequence modeling and demand evolution prediction, providing intelligent reference for design solution iteration.

[0029] S4.2. Combining the three major goals of clothing comfort, production cost, and design aesthetics, the NSGA-II evolutionary algorithm is used for multi-objective optimization scheduling, automatically finding the optimal compromise solution in the objective space and outputting the Pareto frontier set;

[0030] S4.3. Based on digital twin technology, the design plan is mapped to the virtual production workshop. Through discrete event simulation and process parameter modeling, production efficiency and quality risks are predicted to achieve closed-loop verification between design and production.

[0031] S4.4. Utilize IoT sensors and the MES system to monitor data from all stages of sample production in real time, feed actual production deviations back to the generative design engine, and automatically calibrate design parameters through adversarial learning.

[0032] The present invention is further configured as follows: in step S5, in the end-to-end intelligent interactive management supply chain:

[0033] S5.1. Build an intelligent production scheduling system based on graph optimization and reinforcement learning to allocate spatiotemporal resources for equipment such as sewing machines and cutting tables, minimize the total production cycle, and balance the load.

[0034] S5.2. Automatically generate RFID / NFC tag information and combine blockchain distributed ledger technology to record product life cycle data;

[0035] S5.3. Leverage the vehicle routing problem (VRP) and real-time traffic big data to calculate optimal delivery routes through reinforcement learning or cellular automation models, taking into account both carbon emissions and timeliness requirements.

[0036] S5.4. Collect user feedback and return and exchange data after wearing, and continuously update user portraits and demand models through large-scale online learning to provide accurate decision support for the next generation of design and form an end-to-end closed-loop optimization system.

[0037] The present invention is further configured as follows: in the step S3, real-time interactive feedback of three-dimensional fitting:

[0038] S3.1.1. Build personalized 3D body models of users using 3D laser scanning technology, convert clothing designs into a workable FBX file format, and perform geometric simplification, optimization, and texturing on these models;

[0039] S3.1.2. Use the PBR model to set up the interaction between materials and light sources, accurately simulating the reflection effects of clothing under different lighting conditions. HDR (High Dynamic Range) lighting and environment mapping are used to achieve light source, environmental reflection, shadow, gloss and other effects. During the dynamic rendering process, the reflection and refraction effects are calculated in real time to ensure that the gloss and texture of the clothing can be highly restored to the real environment.

[0040] S3.1.3. By integrating real-time ray tracing technology, we can simulate the lighting of clothing surface details (wrinkles, highlights, transparent materials) more realistically, providing more refined shadow, reflection, refraction and other effects. Real-time ray tracing supports the realistic performance of clothing in dynamic scenes and simulates the multiple interactions between light and object surfaces.

[0041] S3.1.4. Use the PhysX fabric physics engine to simulate the dynamic physics of clothing, simulating the behavior of fabric under different movements (such as wrinkles, drapes, and swings). By accurately simulating factors such as fabric elasticity, gravity, wind, and friction, we ensure that the dynamic response of each garment during user movement conforms to the laws of real physics.

[0042] S3.1.5. Integrate motion capture technology to capture the user's motion data in real time (body-worn motion sensors) and drive the avatar's muscle system. Based on the human skeleton and muscle binding model, accurately simulate the impact of muscle changes during user movement on clothing.

[0043] S3.1.6. When the user interacts with the fitting system, the system provides real-time feedback on the changes in clothing under these dynamic behaviors based on the user's input actions (turning, jumping, etc.). The system will optimize the clothing rendering based on the preset judgment branches;

[0044] S3.1.7. Cross-platform rendering is achieved through WebGL and Unity engines to ensure that users can obtain consistent high-fidelity fitting effects whether on PC, mobile phone or virtual reality (VR) devices.

[0045] The present invention is further configured as follows: in step S3.1.6, the action currently input by the user adjusts the calculation accuracy and fabric response of the physical engine according to the fabric type of the clothing and the type of user action. If rapid movement or high-frequency action is detected, the system automatically enables high-precision simulation mode, otherwise it enables low resource consumption mode.

[0046] Beneficial effects

[0047] Compared with the known public technology, the technical solution provided by the present invention has the following beneficial effects:

[0048] 1. The present invention integrates information from different data sources through multi-dimensional user portrait data modeling and fusion to build accurate user portraits. It combines knowledge graphs with spatiotemporal data acquisition technology to obtain environmental context in real time, providing personalized data support for subsequent clothing design. Sentiment analysis and natural language processing technology are used to analyze users' aesthetic tendencies and style preferences, and generate clothing designs that meet user needs based on deep learning models. Through real-time evaluation and optimization of design sketches, it ensures that each design conforms to fashion trends and user preferences. Through personalized recommendations and diversified designs, the innovation and uniqueness of design solutions can be improved.

[0049] 2. The present invention generates a personalized 3D model of the user through 3D scanning technology and point cloud reconstruction, providing accurate dimensional data for clothing design. Combined with PBR and real-time ray tracing technology, the system can simulate reflection, shadow, gloss and other effects under different lighting conditions, presenting extremely realistic clothing effects. Through motion capture and physics engine technology, the user's movements are captured in real time, and the reaction of the fabric under dynamic behavior is simulated to ensure that the clothing conforms to the laws of physics during movement. Through feedback-based real-time optimization and cross-platform rendering, users can obtain a consistent high-fidelity try-on experience on multiple devices such as PCs, mobile phones, and virtual reality devices, thereby greatly improving the user's immersion and customization efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a schematic flow chart of the steps of a method for generating a personalized clothing design scheme based on artificial intelligence of the present invention;

[0051] Figure 2 This is a schematic diagram of the visual interaction judgment branches of the artificial intelligence-based personalized clothing design scheme generation method of the present invention. DETAILED DESCRIPTION

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] The present invention will be further described below with reference to the embodiments.

[0054] Example 1

[0055] like Figure 1 As shown, the present invention provides a technical solution: a method for generating a personalized clothing design solution based on artificial intelligence, comprising the following steps:

[0056] S1. Multi-dimensional user portrait data modeling and integration;

[0057] S2, generative style evolution design engine;

[0058] S3, real-time interactive feedback 3D fitting;

[0059] S4, intelligent iterative optimization and verification;

[0060] S5, end-to-end intelligent interactive management supply chain;

[0061] In the step S1, multi-dimensional user portrait data modeling and fusion:

[0062] S1.1. Utilizing multi-source heterogeneous data fusion and integration technology, we unified the modeling of questionnaires, social media behavior logs, and wearable device physiological indicators through a graph database. We then constructed user attribute nodes based on a knowledge graph algorithm, accurately mapping multi-dimensional portraits such as height, body shape, style, and color preferences.

[0063] S1.2. Utilize spatiotemporal data collection and edge computing to obtain weather, scene lighting, and activity trajectories in real time at the user terminal. Dynamically model the environmental context through a time-series graph neural network and integrate it with the user profile node.

[0064] S1.3. Using the RoBERTa pre-trained large model combined with multimodal emotion recognition, we input the user's natural language description, expressions, and voice intonation into the Transformer emotion classifier to extract aesthetic tendencies and style appeals, generate emotion embedding vectors, and optimize the association with the user profile.

[0065] S1.4. Use structured light and TOF cameras to obtain high-precision depth maps, use point cloud reconstruction and shape restoration algorithms to generate a 3D human mesh model, and perform parameter annotation and correction based on subdivision surfaces and ergonomic standards.

[0066] In the step S2, the generative style evolution design engine:

[0067] S2.1. Based on a convolutional neural network and a multi-head self-attention mechanism, we extract structured style features (silhouettes, patterns, decorative elements, etc.) from massive design galleries and fashion trend data, and dynamically update the style prototype library through a style clustering algorithm.

[0068] S2.2. We build a conditional GAN ​​(cGAN) combined with a variational autoencoder (VAE). This approach uses user profiles and environment vectors as conditional inputs, generates diverse design sketches through a cross-alignment loss and a multi-level discriminator, and evaluates the generated results in real-time using the FID and LPIPS metrics.

[0069] S2.3. Introducing wearable deformation simulation based on finite element analysis (FEA) into the generated garment sketch, automatically adjusting fabric shrinkage, stress distribution, and wrinkle effects through physical constraints and shape optimization iterative algorithms;

[0070] S2.4. Utilizing collaborative filtering and deep matrix factorization algorithms, we integrate user sentiment embedding with popular color models (Pantone trend color library) to output optimal color combinations and fabric material ratios. Through GAN, we generate realistic material maps that can be directly used for virtual rendering or production samples.

[0071] In the step S3, real-time interactive feedback of three-dimensional fitting:

[0072] S3.1, combining PBR (Physically Based Rendering) and real-time ray tracing technology, allows 3D try-on in the browser based on WebGL or Unity engine, supporting fabric gravity, collision and muscle binding simulation, allowing users to intuitively evaluate the wearing effect under different lighting and movements;

[0073] S3.2. Using motion capture or mobile phone inertial navigation data, the virtual human skeleton and fabric physics engine are driven in real time to generate non-static fitting scenes based on the user's daily movements. Spatiotemporal smoothing filtering is also used to ensure smooth system response and natural simulation.

[0074] S3.3, supports multimodal interaction methods such as touch screen gestures, voice and gesture recognition. Users can fine-tune clothing details through voice commands or gestures. The system predicts and optimizes interactive behaviors based on a reinforcement learning strategy network;

[0075] S3.4. Using online A / B testing and the Multi-Armed Bandit algorithm, we dynamically collect user clicks, dwell time, and secondary editing behaviors during the try-on process. We then use Bayesian optimization to calculate satisfaction scores in real time to guide the next round of solution iterations.

[0076] In the step S4, intelligent iterative optimization verification:

[0077] S4.1. Input user fitting, interaction, and satisfaction logs into the Transformer-XL time series neural network to perform behavior sequence modeling and demand evolution prediction, providing intelligent reference for design solution iteration.

[0078] S4.2. Combining the three major goals of clothing comfort, production cost, and design aesthetics, the NSGA-II evolutionary algorithm is used for multi-objective optimization scheduling, automatically finding the optimal compromise solution in the objective space and outputting the Pareto frontier set;

[0079] S4.3. Based on digital twin technology, the design plan is mapped to the virtual production workshop. Through discrete event simulation and process parameter modeling, production efficiency and quality risks are predicted to achieve closed-loop verification between design and production.

[0080] S4.4. Utilize IoT sensors and the MES system to monitor data from all stages of sample production in real time, feed actual production deviations back to the generative design engine, and automatically calibrate design parameters through adversarial learning.

[0081] In the step S5, end-to-end intelligent interactive management supply chain:

[0082] S5.1. Build an intelligent production scheduling system based on graph optimization and reinforcement learning to allocate spatiotemporal resources for equipment such as sewing machines and cutting tables, minimize the total production cycle, and balance the load.

[0083] S5.2. Automatically generate RFID / NFC tag information and combine blockchain distributed ledger technology to record product life cycle data;

[0084] S5.3. Leverage the vehicle routing problem (VRP) and real-time traffic big data to calculate optimal delivery routes through reinforcement learning or cellular automation models, taking into account both carbon emissions and timeliness requirements.

[0085] S5.4. Collect user feedback and return and exchange data after wearing, and continuously update user portraits and demand models through large-scale online learning to provide accurate decision support for the next generation of design and form an end-to-end closed-loop optimization system.

[0086] In this embodiment, multi-dimensional user portrait modeling technology is used to integrate multi-source data from questionnaires, social media, wearable devices, etc. to build an accurate user portrait, and integrate environmental data and sentiment analysis to accurately depict user needs. Then, through the generative design engine, a deep learning model is used to generate clothing designs that meet user preferences, and on this basis, style evolution and color and material optimization are performed. Users can perform three-dimensional virtual try-on through a real-time interactive feedback system. The system combines motion capture and physics engines to achieve dynamic and realistic clothing display. As user interaction and try-on data are continuously collected, the system optimizes the design plan through intelligent iterative optimization algorithms and seamlessly connects with the production process. Finally, the system uses intelligent supply chain management to achieve rapid production and accurate delivery of personalized clothing, forming an end-to-end closed loop from design to delivery, greatly improving personalized customization efficiency and customer experience.

[0087] Example 2

[0088] like Figure 1-2 As shown, the present invention provides a technical solution: a method for generating personalized clothing design solutions based on artificial intelligence, wherein in step S3, real-time interactive feedback of three-dimensional fitting:

[0089] S3.1.1. Build personalized 3D body models of users using 3D laser scanning technology, convert clothing designs into a workable FBX file format, and perform geometric simplification, optimization, and texturing on these models;

[0090] S3.1.2. Use the PBR model to set up the interaction between materials and light sources, accurately simulating the reflection effects of clothing under different lighting conditions. HDR (High Dynamic Range) lighting and environment mapping are used to achieve light source, environmental reflection, shadow, gloss and other effects. During the dynamic rendering process, the reflection and refraction effects are calculated in real time to ensure that the gloss and texture of the clothing can be highly restored to the real environment.

[0091] S3.1.3. By integrating real-time ray tracing technology, we can simulate the lighting of clothing surface details (wrinkles, highlights, transparent materials) more realistically, providing more refined shadow, reflection, refraction and other effects. Real-time ray tracing supports the realistic performance of clothing in dynamic scenes and simulates the multiple interactions between light and object surfaces.

[0092] S3.1.4. Use the PhysX fabric physics engine to simulate the dynamic physics of clothing, simulating the behavior of fabric under different movements (such as wrinkles, drapes, and swings). By accurately simulating factors such as fabric elasticity, gravity, wind, and friction, we ensure that the dynamic response of each garment during user movement conforms to the laws of real physics.

[0093] S3.1.5. Integrate motion capture technology to capture the user's motion data in real time (body-worn motion sensors) and drive the avatar's muscle system. Based on the human skeleton and muscle binding model, accurately simulate the impact of muscle changes during user movement on clothing.

[0094] S3.1.6. When the user interacts with the fitting system, the system provides real-time feedback on the changes in clothing under these dynamic behaviors based on the user's input actions (turning, jumping, etc.). The system will optimize the clothing rendering based on the preset judgment branches;

[0095] S3.1.7. Cross-platform rendering is achieved through WebGL and Unity engines, ensuring that users can obtain consistent high-fidelity fitting effects regardless of whether they are on PC, mobile phone, or virtual reality (VR) devices.

[0096] In step S3.1.6, the action currently input by the user adjusts the calculation accuracy and fabric response of the physical engine according to the fabric type of the clothing and the type of user action. If fast movement or high-frequency action is detected, the system automatically enables high-precision simulation mode, otherwise it enables low resource consumption mode.

[0097] In this embodiment, 3D laser scanning technology is used to construct a personalized 3D body model of the user, and the clothing design is converted into an operable 3D file format for geometric simplification and optimization. The PBR (Physically Based Rendering) model is used to interactively set the clothing material and light source, simulating reflection, shadow, gloss and other effects under different lighting conditions, ensuring that the clothing highly reproduces the real lighting environment during dynamic rendering. At the same time, real-time ray tracing technology is integrated to finely simulate the light reflection and refraction of clothing surface details, further enhancing the realism of the clothing. In terms of dynamic simulation, the system uses the PhysX cloth physics engine to simulate the behavior of cloth under different movements, accurately reflecting effects such as wrinkles, drapes, and swings, ensuring that the dynamic response of the clothing conforms to the laws of real physics. Motion capture technology is used to collect user movement data in real time and drive the avatar's muscle system to accurately simulate the changes of clothing during movement. The system adjusts the calculation accuracy of the physics engine in real time based on the user's dynamic input, such as turning or jumping, to optimize the rendering effect of the clothing, ensure consistency across different devices (such as PCs, mobile phones or VR devices), and provide an immersive and high-fidelity try-on experience.

[0098] Working principle:

[0099] like Figure 1-2 As shown in the figure, in actual use, through multi-dimensional user portrait data modeling and fusion, information from different data sources (such as questionnaires, social media behavior logs, wearable devices, etc.) is integrated to build accurate user portraits. The knowledge graph algorithm and spatiotemporal data acquisition technology are combined with edge computing to obtain environmental context information (such as weather, lighting, etc.) in real time, providing personalized data support for subsequent clothing design. Sentiment analysis and natural language processing technology are also used to analyze the user's aesthetic and style preferences, further optimize the design results, and through 3D scanning technology and point cloud reconstruction, the user's body size data is accurately obtained, providing a basis for the design and generation of personalized clothing.

[0100] The system utilizes a generative style evolution design engine to automatically generate clothing designs tailored to individual user needs. Combining convolutional neural networks with a multi-head self-attention mechanism, it extracts clothing style features from a vast fashion database and generates diverse design sketches through a conditional GAN ​​combined with VAE. The system continuously updates its style prototype library through real-time evaluation and optimization of design results, ensuring the innovative and fashionable nature of the designs. The selection of fabrics, colors, and materials relies on collaborative filtering and deep learning models. The system automatically recommends color and material combinations that meet the user's preferences, ensuring that each design achieves optimal visual and tactile effects.

[0101] Users interact with the design plan through the real-time interactive feedback 3D fitting system. Combining PBR with real-time ray tracing technology, the system can realistically present clothing effects under different lighting environments. The user's actions (such as turning, jumping, etc.) are captured in real time through motion capture technology. The system adjusts the calculation accuracy of the physical engine according to the user's dynamic behavior, simulates the dynamic response of the fabric and the changes in clothing. During the fitting process, the system performs real-time optimization based on user feedback (such as clicks, dwell time, etc.) to ensure that the clothing effect in the virtual fitting process is real and natural. The use of cross-platform rendering technology ensures that users can get a consistent high-fidelity fitting experience on PCs, mobile phones or virtual reality devices. Through the combination of these technologies, the entire clothing design and fitting process has become more intelligent and personalized, greatly improving the user experience and customization efficiency.

[0102] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for generating personalized clothing design schemes based on artificial intelligence, characterized in that: The following steps are involved: S1. Multi-dimensional user portrait data modeling and integration; S2, generative style evolution design engine; S3, real-time interactive feedback 3D fitting; S4, intelligent iterative optimization and verification; S5. End-to-end intelligent interactive management of the supply chain.

2. The method for generating personalized clothing design schemes based on artificial intelligence according to claim 1, characterized in that: In the step S1, multi-dimensional user portrait data modeling and fusion: S1.

1. Utilizing multi-source heterogeneous data fusion and integration technology, we unified the modeling of questionnaires, social media behavior logs, and wearable device physiological indicators through a graph database. We then constructed user attribute nodes based on a knowledge graph algorithm, accurately mapping multi-dimensional portraits such as height, body shape, style, and color preferences. S1.

2. Utilize spatiotemporal data collection and edge computing to obtain weather, scene lighting, and activity trajectories in real time at the user terminal. Dynamically model the environmental context through a time-series graph neural network and integrate it with the user profile node. S1.

3. Using the RoBERTa pre-trained large model combined with multimodal emotion recognition, we input the user's natural language description, expressions, and voice intonation into the Transformer emotion classifier to extract aesthetic tendencies and style appeals, generate emotion embedding vectors, and optimize the association with the user profile. S1.

4. Use structured light and TOF cameras to obtain high-precision depth maps, use point cloud reconstruction and shape repair algorithms to generate a 3D human mesh model, and perform parameter annotation and correction based on subdivision surfaces and ergonomic standards.

3. The method for generating personalized clothing design schemes based on artificial intelligence according to claim 1, characterized in that: In the step S2, the generative style evolution design engine: S2.

1. Based on a convolutional neural network and a multi-head self-attention mechanism, we extract structured style features (silhouettes, patterns, decorative elements, etc.) from massive design galleries and fashion trend data, and dynamically update the style prototype library through a style clustering algorithm. S2.

2. We build a conditional GAN ​​(cGAN) combined with a variational autoencoder (VAE). This approach uses user profiles and environment vectors as conditional inputs, generates diverse design sketches through a cross-alignment loss and a multi-level discriminator, and evaluates the generated results in real-time using the FID and LPIPS metrics. S2.

3. Introducing wearable deformation simulation based on finite element analysis (FEA) into the generated garment sketch, automatically adjusting fabric shrinkage, stress distribution, and wrinkle effects through physical constraints and shape optimization iterative algorithms; S2.

4. Utilize collaborative filtering and deep matrix decomposition algorithms to integrate user sentiment embedding with popular color models (Pantone trend color library), output the optimal color combination and fabric material ratio, and generate realistic material maps through GAN, which can be directly used for virtual rendering or production samples.

4. The method for generating personalized clothing design solutions based on artificial intelligence according to claim 1, wherein: In the step S3, real-time interactive feedback of three-dimensional fitting: S3.1, combining PBR (Physically Based Rendering) and real-time ray tracing technology, allows 3D try-on in the browser based on WebGL or Unity engine, supporting fabric gravity, collision and muscle binding simulation, allowing users to intuitively evaluate the wearing effect under different lighting and movements; S3.

2. Using motion capture or mobile phone inertial navigation data, the virtual human skeleton and fabric physics engine are driven in real time to generate non-static fitting scenes based on the user's daily movements. Spatiotemporal smoothing filtering is also used to ensure smooth system response and natural simulation. S3.3, supports multimodal interaction methods such as touch screen gestures, voice and gesture recognition. Users can fine-tune clothing details through voice commands or gestures. The system predicts and optimizes interactive behaviors based on a reinforcement learning strategy network; S3.

4. Use online A / B testing and multi-armed bandit algorithms (Multi-ArmedBandit), dynamically collects user clicks, dwell time, and secondary editing behaviors during the fitting process, and calculates satisfaction scores in real time through Bayesian optimization to guide the next round of solution iteration.

5. The method for generating personalized clothing design schemes based on artificial intelligence according to claim 1, characterized in that: In the step S4, intelligent iterative optimization verification: S4.

1. Input user fitting, interaction, and satisfaction logs into the Transformer-XL time series neural network to perform behavior sequence modeling and demand evolution prediction, providing intelligent reference for design solution iteration. S4.

2. Combining the three major goals of clothing comfort, production cost, and design aesthetics, the NSGA-II evolutionary algorithm is used for multi-objective optimization scheduling, automatically finding the optimal compromise solution in the objective space and outputting the Pareto frontier set; S4.

3. Based on digital twin technology, the design plan is mapped to the virtual production workshop. Through discrete event simulation and process parameter modeling, production efficiency and quality risks are predicted to achieve closed-loop verification between design and production. S4.

4. Utilize IoT sensors and the MES system to monitor data from all stages of sample production in real time, feed actual production deviations back to the generative design engine, and automatically calibrate design parameters through adversarial learning.

6. The method for generating personalized clothing design scheme based on artificial intelligence according to claim 1, characterized in that: In the step S5, end-to-end intelligent interactive management supply chain: S5.

1. Build an intelligent production scheduling system based on graph optimization and reinforcement learning to allocate spatiotemporal resources for equipment such as sewing machines and cutting tables, minimize the total production cycle, and balance the load. S5.

2. Automatically generate RFID / NFC tag information and combine blockchain distributed ledger technology to record product life cycle data; S5.

3. Leverage the vehicle routing problem (VRP) and real-time traffic big data to calculate optimal delivery routes through reinforcement learning or cellular automation models, taking into account both carbon emissions and timeliness requirements. S5.

4. Collect user feedback and return and exchange data after wearing, and continuously update user portraits and demand models through large-scale online learning to provide accurate decision support for the next generation of design and form an end-to-end closed-loop optimization system.

7. The method for generating personalized clothing design scheme based on artificial intelligence according to claim 4, characterized in that: In the step S3, real-time interactive feedback of three-dimensional fitting: S3.1.

1. Build personalized 3D body models of users using 3D laser scanning technology, convert clothing designs into a workable FBX file format, and perform geometric simplification, optimization, and texturing on these models; S3.1.

2. Use the PBR model to set up the interaction between materials and light sources, accurately simulating the reflection effects of clothing under different lighting conditions. HDR (High Dynamic Range) lighting and environment mapping are used to achieve light source, environmental reflection, shadow, gloss and other effects. During the dynamic rendering process, the reflection and refraction effects are calculated in real time to ensure that the gloss and texture of the clothing can be highly restored to the real environment. S3.1.

3. By integrating real-time ray tracing technology, we can simulate the lighting of clothing surface details (wrinkles, highlights, transparent materials) more realistically, providing more refined shadow, reflection, refraction and other effects. Real-time ray tracing supports the realistic performance of clothing in dynamic scenes and simulates the multiple interactions between light and object surfaces. S3.1.

4. Use the PhysX fabric physics engine to simulate the dynamic physics of clothing, simulating the behavior of fabric under different movements (such as wrinkles, drapes, and swings). By accurately simulating factors such as fabric elasticity, gravity, wind, and friction, we ensure that the dynamic response of each garment during user movement conforms to the laws of real physics. S3.1.

5. Integrate motion capture technology to capture the user's motion data in real time (body-worn motion sensors) and drive the avatar's muscle system. Based on the human skeleton and muscle binding model, accurately simulate the impact of muscle changes during user movement on clothing. S3.1.

6. When the user interacts with the fitting system, the system provides real-time feedback on the changes in clothing under these dynamic behaviors based on the user's input actions (turning, jumping, etc.). The system will optimize the clothing rendering based on the preset judgment branches; S3.1.

7. Cross-platform rendering is achieved through WebGL and Unity engines to ensure that users can obtain consistent high-fidelity fitting effects whether on PC, mobile phone or virtual reality (VR) devices.

8. The method for generating personalized clothing design scheme based on artificial intelligence according to claim 7, characterized in that: In step S3.1.6, the action currently input by the user adjusts the calculation accuracy and fabric response of the physical engine according to the fabric type of the clothing and the type of user action. If fast movement or high-frequency action is detected, the system automatically enables high-precision simulation mode, otherwise it enables low resource consumption mode.

Images