Garment design system and method based on artificial intelligence

Abstract

The invention provides a costume design system and method based on artificial intelligence, and relates to the field of artificial intelligence, and the method comprises the steps: collecting multi-dimensional design data through a cloud collaborative architecture of costume design resources; performing dynamic hierarchical allocation on the multi-dimensional design data to obtain a hierarchical allocation result, and performing sparse processing on the multi-dimensional design data to obtain a sparse model of the costume design resources; performing strategy training based on deep reinforcement learning on the sparse model according to a hierarchical distribution result and historical collaborative data to obtain a sharing strategy and an optimized migration parameter; constructing a knowledge migration component of the cloud collaborative architecture based on the sharing strategy and the optimized migration parameters; knowledge migration optimization is carried out on the collaborative process of the costume design resources based on the knowledge migration component through the cloud collaborative architecture, the costume shared resources for intelligent generalization are generated, dynamic hierarchical distribution and knowledge migration optimization can be carried out on the costume design resources, and intelligent generalization sharing of the costume design resources is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, and more particularly, to a clothing design system and method based on artificial intelligence. BACKGROUND

[0002] Artificial intelligence can allocate computing, storage and network resources on demand, effectively support large-scale data processing and intelligent collaboration of multi-terminal devices, and become an important technical foundation for supporting the digital transformation of modern enterprises. In the field of design, with the increasing complexity of design tasks and collaboration needs, intelligent collaborative sharing methods of clothing design resources based on cloud services have gradually attracted attention.

[0003] Clothing design based on artificial intelligence can combine edge computing and real-time interaction of terminal devices to realize cross-regional sharing and distributed collaborative operation of clothing design resources. However, the existing intelligent collaborative sharing method of clothing design resources based on cloud services usually adopts a static hierarchical or fixed resource allocation strategy, lacks support for dynamic hierarchical allocation of clothing design resources, and cannot effectively cope with the diversification and multi-dimensional characteristics of design tasks. In addition, relying only on static models or a single training strategy, it is difficult to achieve intelligent sharing and migration. In addition, for the multi-terminal collaboration needs in a distributed environment, the existing method still has deficiencies in the collection, sparsification processing and dynamic optimization of multi-dimensional design data. Therefore, how to dynamically allocate and optimize knowledge migration of clothing design resources to realize intelligent generalization sharing of clothing design resources is a difficult problem in the industry. SUMMARY

[0004] The present application provides a clothing design system and method based on artificial intelligence, which can dynamically allocate and optimize knowledge migration of clothing design resources to realize intelligent generalization sharing of clothing design resources.

[0005] In a first aspect, the present application provides a clothing design method based on artificial intelligence, which comprises the following steps: determining a cloud collaborative architecture of clothing design resources, and then collecting multi-dimensional design data through the cloud collaborative architecture; dynamically allocating the multi-dimensional design data to obtain a hierarchical allocation result of the clothing design resources, and sparsifying the multi-dimensional design data to obtain a sparse model of the clothing design resources; obtaining historical collaborative data of the clothing design resources, performing policy training on the sparse model based on deep reinforcement learning according to the hierarchical allocation result and the historical collaborative data, and then obtaining a sharing strategy and an optimized migration parameter of the clothing design resources; constructing a knowledge migration component of the cloud collaborative architecture based on the sharing strategy and the optimized migration parameter; The cloud collaboration architecture optimizes knowledge migration of a collaborative process of a clothing design resource based on the knowledge migration component, and generates a clothing sharing resource for intelligent generalization.

[0006] In this embodiment, the cloud collaboration architecture includes a cloud, an edge layer, and a terminal layer.

[0007] In this embodiment, multi-dimensional design data is collected through the terminal layer in the cloud collaboration architecture, and the terminal layer includes a plurality of design terminals for performing design tasks.

[0008] In this embodiment, the dynamic hierarchical allocation of the multi-dimensional design data obtains a hierarchical allocation result of the clothing design resource, which specifically includes: determining a multi-dimensional index set of the multi-dimensional design data; dynamically clustering the multi-dimensional design data based on the multi-dimensional index set to obtain different clustering clusters; hierarchically allocating all clustering clusters to obtain a hierarchical allocation result of the clothing design resource.

[0009] In this embodiment, the multi-dimensional design data is processed by sparse processing to obtain a sparse model of the clothing design resource, which specifically includes: performing an unstructured pruning operation on the preprocessed multi-dimensional design data to obtain pruned data; determining a sparse ratio of the clothing design resource based on a preset mask matrix and the pruned data; adaptively adjusting the sparse ratio to obtain a sparse model of the clothing design resource.

[0010] In this embodiment, the historical collaboration data of the clothing design resource is obtained through the operation log of the design terminal in the cloud collaboration architecture.

[0011] In this embodiment, the sparse model is trained based on deep reinforcement learning according to the hierarchical allocation result and the historical collaboration data, and further obtains a sharing strategy and an optimized migration parameter of the clothing design resource, which specifically includes: constructing a hierarchical reward function through the hierarchical allocation result; training the sparse model based on the historical collaboration data through the hierarchical reward function to obtain an optimized migration parameter of the clothing design resource; adopting a deep deterministic policy gradient algorithm to perform policy estimation on the sparse model based on the historical collaboration data to obtain a sharing strategy of the clothing design resource.

[0012] In this embodiment, the knowledge migration component of the cloud collaboration architecture is constructed based on the sharing strategy and the optimized migration parameter, which specifically includes: initialize the knowledge transfer component architecture, and load the shared strategy and the optimization transfer parameter through the knowledge transfer component architecture; The optimization transfer parameter performs adaptive optimization on the generalization ability of the knowledge transfer component architecture according to a learning rate, a weight decay, and a pruning ratio, to obtain an optimized knowledge transfer component architecture. Based on the shared strategy, the optimized knowledge transfer component architecture is responded layer by layer, and when the response rate of the shared strategy reaches a preset index, the optimized knowledge transfer component architecture is taken as the knowledge transfer component of the cloud collaborative architecture.

[0013] In the embodiment, the knowledge transfer component is a multi-level transfer learning unit based on federated transfer learning.

[0014] In a second aspect, the present application provides an artificial intelligence-based clothing design system for executing an artificial intelligence-based clothing design method, the clothing design system comprising: A multi-dimensional collaboration module is configured to determine a cloud collaborative architecture of clothing design resources, and to collect multi-dimensional design data through the cloud collaborative architecture. A hierarchical and sparse module is configured to dynamically allocate the multi-dimensional design data to obtain a hierarchical allocation result of the clothing design resources, and to perform sparse processing on the multi-dimensional design data to obtain a sparse model of the clothing design resources. A strategy training module is configured to obtain historical collaboration data of the clothing design resources, to perform strategy training on the sparse model based on deep reinforcement learning according to the hierarchical allocation result and the historical collaboration data, and to obtain a shared strategy and an optimization transfer parameter of the clothing design resources. A component construction module is configured to construct a knowledge transfer component of the cloud collaborative architecture based on the shared strategy and the optimization transfer parameter. A shared optimization module is configured to perform knowledge transfer optimization on the collaboration process of the clothing design resources based on the knowledge transfer component of the cloud collaborative architecture, to generate clothing shared resources for intelligent generalization.

[0015] The technical scheme provided by the embodiments of the present application has the following beneficial effects: The cloud collaboration architecture of the clothing design resource is determined, and multi-dimensional design data is collected through the cloud collaboration architecture. The multi-dimensional design data is dynamically layered and distributed to obtain a layered distribution result of the clothing design resource, and the multi-dimensional design data is sparsified to obtain a sparse model of the clothing design resource. Historical collaboration data of the clothing design resource is obtained, the sparse model is trained based on deep reinforcement learning according to the layered distribution result and the historical collaboration data, and a sharing strategy and an optimized migration parameter of the clothing design resource are obtained. A knowledge migration component of the cloud collaboration architecture is constructed based on the sharing strategy and the optimized migration parameter. The cloud collaboration architecture optimizes the collaboration process of the clothing design resource based on the knowledge migration component, and generates clothing sharing resources for intelligent generalization.

[0016] It can be seen that in the present application, the clothing design resource can be dynamically layered and distributed and knowledge migration optimized. First, the cloud collaboration architecture of the clothing design resource realizes efficient collaboration of multi-terminal, edge and cloud, unifies the storage, distribution and management system of the clothing design resource, ensures efficient collection and unified access of multi-dimensional design data, and dynamically layers and distributes the multi-dimensional design data, adaptively divides the clothing design resource into different layered structures, and facilitates on-demand distribution and task matching of the clothing design resource. At the same time, through the sparsification processing, the training efficiency and sharing ability of the model can be improved on the basis of preserving the key features of the clothing design resource, thereby providing a structured input model for subsequent knowledge migration optimization. Second, the sparse model is trained by deep reinforcement learning, which can fully exploit the collaboration mode and historical data rules of the clothing design resource, and is beneficial to intelligent extraction of the clothing design resource sharing strategy and optimized migration parameter. Then, by constructing the knowledge migration component of the cloud collaboration architecture, the sharing demand of the clothing design resource can be responded layer by layer based on the learning rate, weight decay and pruning ratio, the dynamic optimization and structured adaptation of the knowledge migration component are realized, and the intelligent generalization sharing ability of the clothing design resource in cross-terminal and cross-scene is further enhanced. Finally, the cloud collaboration architecture optimizes the clothing design resource based on the knowledge migration component, which can fully exploit the historical knowledge and dynamic layered features of the clothing design resource, realize intelligent migration and adaptive generalization of the resource, and generate clothing design resources with intelligent generalization ability according to actual design task requirements, which is beneficial to improving the cross-domain sharing efficiency of the clothing design resource and the intelligent level of the system as a whole.

[0017] In summary, the technical solution adopted in the present application can dynamically layer and distribute the clothing design resource and optimize the knowledge migration, to realize intelligent generalization and sharing of the clothing design resource. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description only represent some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort.

[0019] Figure 1 is a flowchart of a clothing design method based on artificial intelligence provided by the present application; Figure 2 is an exemplary flowchart for determining the hierarchical allocation result of clothing design resources provided by the present application; Figure 3 is an exemplary flowchart for determining the sharing strategy and optimization migration parameters of clothing design resources provided by the present application; Figure 4 is a module structure diagram of a clothing design system based on artificial intelligence provided by the present application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments only represent some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort are within the scope of protection of the present application.

[0021] The present application provides a clothing design system and method based on artificial intelligence, which determines a cloud collaborative architecture of clothing design resources, and then collects multi-dimensional design data through the cloud collaborative architecture. The multi-dimensional design data is dynamically allocated in layers to obtain a hierarchical allocation result of clothing design resources, and the multi-dimensional design data is processed by sparsification to obtain a sparse model of clothing design resources. Historical collaborative data of clothing design resources is obtained, and the sparse model is trained based on deep reinforcement learning according to the hierarchical allocation result and the historical collaborative data, and then a sharing strategy and optimization migration parameters of clothing design resources are obtained. The knowledge migration component of the cloud collaborative architecture is constructed based on the sharing strategy and the optimization migration parameters. The cloud collaborative architecture optimizes the collaborative process of clothing design resources based on the knowledge migration component, and generates clothing sharing resources for intelligent generalization.

[0022] Embodiment one, in order to better understand the above technical solutions, the above technical solutions will be described in detail in the following with reference to the drawings in the specification and specific embodiments. Refer to Figure 1As shown in the figure, the figure is a flowchart of a clothing design method based on artificial intelligence according to the embodiment of the present application, and the clothing design method comprises the following steps: In step S1, the cloud collaboration architecture of the clothing design resource is determined, and then the multi-dimensional design data is collected through the cloud collaboration architecture.

[0023] It should be noted that the cloud collaboration architecture in the present application includes: cloud, edge layer and terminal layer; wherein the cloud layer is used for unified management, storage and distribution of clothing design resources, and can centrally manage multi-dimensional design data; the edge layer is used for preliminary processing of design tasks, including data preprocessing, caching and security encryption; preferably, in the present application, the edge layer deploys a neural network, which can automatically extract features and semantically annotate clothing design resources through the neural network. The neural network can use a lightweight U-Net, which is beneficial to fast retrieval, preliminary optimization and secure sharing of clothing design resources, so as to ensure the integrity and security of the data before uploading to the cloud; the terminal layer includes a plurality of design terminals, and the design terminal is used for executing the design task. The design terminal can include PC terminal, mobile terminal and design software plug-in, etc.

[0024] In specific implementation, the multi-dimensional design data is collected through the terminal layer of the cloud collaboration architecture, wherein the multi-dimensional design data includes: design model file, design parameter, design process log and user operation record, etc. The collected multi-dimensional design data can be preliminarily processed through the edge layer and then uploaded to the cloud for unified management through data transmission technology, which is convenient for subsequent analysis and processing.

[0025] In step S2, the multi-dimensional design data is dynamically layered and distributed to obtain the layered distribution result of the clothing design resource, and the multi-dimensional design data is sparsified to obtain the sparse model of the clothing design resource.

[0026] Preferably, in the present embodiment, reference Figure 2 As shown in the figure, the figure is an exemplary flowchart for determining the layered distribution result of the clothing design resource according to the present application. In the present embodiment, the multi-dimensional design data is dynamically layered and distributed to obtain the layered distribution result of the clothing design resource, which can be realized by the following steps: First, in step S21, the multi-dimensional index set of the multi-dimensional design data is determined; Then, in step S22, the multi-dimensional design data is dynamically clustered based on the multi-dimensional index set to obtain different clustering clusters; Finally, in step S23, all clustering clusters are layered to obtain the layered distribution result of the clothing design resource.

[0027] In practical implementation, firstly, feature extraction and standardization are performed on the multidimensional design data. This includes: extracting structural features such as file size and model complexity (e.g., number of grids, number of topological layers) from the design model file; extracting statistical features such as data dimension, mean and variance of data distribution, and missing data rate from the design parameter data; extracting operation frequency, operation sequence information, and user operation behavior features from the design process log; and extracting task features such as task priority, task category, and task execution status from the task context information. Preferably, the above feature extraction and standardization process can be performed using the `pand` utility in Python. The system implements the above features using the AS library, performs unified normalization on the features, and uses the set of normalized features as a multi-dimensional index set. Then, based on the multi-dimensional index set, the dynamic clustering algorithm K-means is used to dynamically cluster the multi-dimensional index vectors, grouping similar design data into the same cluster, thus obtaining different clusters. Finally, based on existing indicators such as task priority, resource requirements, and data dependencies, each cluster is hierarchically divided. Preferably, a tree-like hierarchical algorithm can be used to divide each cluster into layers, which can obtain the hierarchical allocation results of clothing design resources and reduce the failure rate of static hierarchical methods under dynamic task changes.

[0028] It should be noted that, in this application, the hierarchical allocation results of clothing design resources are information used to guide resource sparsity modeling, knowledge transfer optimization, and strategy generation. These hierarchical allocation results include: resource level labels, clustering affiliation information, task adaptation levels, inter-layer dependencies, and resource scheduling priorities. Specifically, resource level labels can be hierarchical identifiers for each type of clothing design resource allocation, such as: basic structure layer, general template layer, and high-frequency reuse layer, used to distinguish the importance and reuse frequency of resources in the system; clustering affiliation information is used to represent their feature similarity; task adaptation levels indicate the type, complexity, or priority of design tasks that the resource at that level is adapted to; inter-layer dependencies define the calling order, dependency direction, or logical constraints between levels; and resource scheduling priorities are used to determine the calling order during resource allocation and sharing, and are optimized in conjunction with task scheduling strategies. Furthermore, the multi-dimensional indicator set, through the extraction and unified normalization of structural features, statistical features, operational behavior features, and task features, ensures the comparability and analyzability of data across different dimensions and sources.

[0029] In this embodiment, the sparse processing of the multidimensional design data to obtain a sparse model of clothing design resources can be carried out in the following manner: After preprocessing the multidimensional design data, an unstructured pruning operation is performed to obtain pruning data; The sparsity ratio of the clothing design resources is determined based on the preset mask matrix and the pruning data. The sparsity ratio is adaptively adjusted to obtain a sparsity model for clothing design resources.

[0030] In specific implementation, firstly, the multidimensional design data undergoes format unification, missing value processing, and standardization to ensure consistency across different sources and dimensions. Then, a gradient pruning algorithm is used to perform unstructured pruning on the preprocessed multidimensional design data. This unstructured pruning includes: calculating the gradient contribution value of each data unit in the multidimensional design data within the model; filtering data sequences with gradient contribution values ​​less than a preset threshold based on a preset threshold; and removing zero values ​​from the data sequences to obtain pruned data. Next, based on a preset mask matrix, the pruned data is modeled as sparsity. This preset mask matrix records the index information of the pruned data units and can be preset using a binary matrix. Finally, a neural network model is initialized, and the sparsity ratio is adjusted based on the performance metrics of the validation set (e.g., accuracy, loss value). When the slope of the sparsity ratio change curve is 0 (i.e., the sparsity ratio change curve is a horizontal straight line parallel to the x-axis), this neural network model is used as the sparsity model for the clothing design resources. The validation set can be constructed using historical collaborative data of the clothing design resources.

[0031] It should be noted that the sparsity model in this application refers to a neural network model with low redundancy generated after pruning, compressing, and structurally diluting multidimensional design data. In this application, structured pruning operations can reduce the computational and storage overhead of the model while preserving its core performance. The mask matrix is ​​used to record the index information of the data units retained after pruning, and can be preset using a binary matrix to ensure the correctness of the model during inference and training. The sparsity ratio refers to the proportion of retained data units in the model after sparsification to the original data units, and is an important indicator for measuring the degree of sparsity. Adaptive adjustment refers to the process of dynamically adjusting the sparsity ratio based on model performance indicators. By setting a validation set and conducting multiple iterative training, the adaptability of the sparsity model under different task scenarios can be improved.

[0032] In step S3, historical collaborative data of clothing design resources is obtained, and the sparsification model is trained based on deep reinforcement learning strategy according to the hierarchical allocation results and the historical collaborative data, so as to obtain the sharing strategy and optimized migration parameters of clothing design resources.

[0033] It should be noted that, in this embodiment, historical collaborative data of clothing design resources is obtained through the operation logs of multiple design terminals in the cloud collaborative architecture. The historical collaborative data refers to the historical operation data generated and recorded by the design terminals during the collaborative sharing of clothing design resources. It can reflect the execution status and collaborative relationship of the design tasks of multiple terminals and is used to provide data support for subsequent deep reinforcement learning training. Optionally, the historical collaborative data can be the historical operation data during the collaboration of the past three design tasks, or it can be set to any other time period according to actual needs.

[0034] In practice, historical collaborative data is collected through the operation logs of the design terminal in the cloud-based collaborative architecture, and can be obtained from the cloud database. This historical collaborative data includes: operation log data of design tasks, resource call records, collaborative behavior data, and task performance indicators. Operation log data includes: the start time, completion time, operation steps, and task execution status of the design task; resource call records include: the number of times the design model file is loaded, its usage frequency, and modification records; collaborative behavior data includes: task allocation, inter-task dependencies, task priorities, multi-person collaboration interaction records, and change history; and task performance indicators include: design task completion efficiency, design data processing latency, and task quality indicators.

[0035] Preferably, in this embodiment, reference Figure 3 As shown, this diagram is an exemplary flowchart for determining the sharing strategy and optimized migration parameters of clothing design resources according to the present application. In this embodiment, the sparse model is trained using a deep reinforcement learning strategy based on the hierarchical allocation results and the historical collaborative data to obtain the sharing strategy and optimized migration parameters of clothing design resources. This can be achieved through the following steps: In step S31, a hierarchical reward function is constructed based on the hierarchical allocation results; In step S32, the hierarchical reward function trains the sparse model on the collaborative efficiency based on the historical collaborative data to obtain the optimized migration parameters of clothing design resources; In step S33, a deep deterministic policy gradient algorithm is used to estimate the policy of the sparse model based on the historical collaborative data to obtain the sharing policy of clothing design resources.

[0036] In specific implementation, firstly, a hierarchical reward function is constructed based on existing indicators such as task priority, resource utilization, and hierarchical dependency relationships through hierarchical allocation results. This hierarchical reward function includes single-layer reward values ​​and cross-layer reward weights, used to score the model's performance under different hierarchical tasks. Then, historical collaborative data and the sparse model are fused, and information such as historical task execution status, operation frequency, and resource call patterns are used to train the sparse model for collaborative efficiency. Preferably, a deep deterministic policy gradient algorithm based on the Actor-Critic architecture can be used for collaborative efficiency training. The Actor-Critic architecture is a reinforcement learning algorithm framework based on policy optimization, including: an Actor network using the state information of the sparse model (including the current level's features)... The model takes eigenvectors, task dependencies, and sparsity ratios as inputs and outputs a sharing strategy for clothing design resources. The Critic network takes the strategy output by the Actor and historical collaborative data as inputs, calculates the value function of the strategy using a hierarchical reward function, and updates the strategy of the Actor network through a backpropagation algorithm. During training, an experience replay mechanism is used to input historical collaborative data in segments according to time windows to enhance the stability and generalization ability of the model. Finally, when the training process converges, the strategy result output by the Actor network is used as the sharing strategy for clothing design resources, and optimization transfer parameters are extracted from the weights of the Critic network. It should be noted that the optimization transfer parameters and the sharing strategy in this application are determined synchronously, that is, steps S32 and S33 are performed synchronously.

[0037] It should be noted that the optimized transfer parameters in this application refer to learnable adaptation factors during reinforcement learning, including learning rate, weight decay, and pruning ratio. Shared policy refers to the policy function used by all agents in a multi-agent reinforcement learning environment. The hierarchical reward function assigns different reward values ​​to the model's performance at different levels, reflecting the quality of the model's policies across multiple tasks. Collaborative efficiency training can combine historical collaborative data (including task states, operation frequencies, and resource call patterns) with a sparse model to specifically improve the efficiency of the model's policies, better adapting to real-world collaborative scenarios. Furthermore, the Deep Deterministic Policy Gradient (DDPG) algorithm is a deep reinforcement learning algorithm based on an Actor-Critic architecture, suitable for policy training in continuous action spaces. It possesses strong policy optimization and generalization capabilities, which is beneficial for improving the sharing efficiency of clothing design resources and the model's adaptive transfer capabilities.

[0038] In step S4, a knowledge migration component of the cloud-based collaborative architecture is constructed based on the sharing strategy and the optimized migration parameters.

[0039] It should be noted that the knowledge transfer component in this application is a multi-level transfer learning unit based on federated transfer learning. It is a functional module that can be used to realize the sharing and optimization transfer of clothing design resources. It is responsible for applying the sharing strategy and optimization transfer parameters obtained from deep reinforcement learning training to the actual sharing process of clothing design resources. In addition, federated transfer learning refers to the transfer and fusion of local model knowledge scattered on various design terminals to the cloud through a multi-terminal distributed collaborative learning framework, which can improve the generalization ability and sharing efficiency of the model in multiple scenarios. The multi-level transfer learning unit refers to the transfer learning module divided into multiple levels (such as the hierarchical structure of clothing design resources, task priority level, and distributed architecture level) within the knowledge transfer component, which is conducive to the efficient intelligent sharing of resources in complex multi-dimensional clothing design resource scenarios.

[0040] In this embodiment, the knowledge migration component for constructing the cloud-based collaborative architecture based on the sharing strategy and the optimized migration parameters can be implemented in the following manner: Initialize the knowledge migration component architecture, and load the sharing strategy and the optimized migration parameters through the knowledge migration component architecture; The optimized transfer parameters are based on the learning rate, weight decay, and pruning ratio to adaptively optimize the generalization ability of the knowledge transfer component architecture, resulting in the optimized knowledge transfer component architecture. Based on the sharing strategy, the optimized knowledge migration component architecture is responded to layer by layer. When the response rate of the sharing strategy reaches the preset index, the optimized knowledge migration component architecture is used as the knowledge migration component of the cloud collaborative architecture.

[0041] In specific implementation, firstly, the knowledge transfer component architecture is initialized. This architecture includes an input layer, a transfer module layer, and an output layer. The input layer receives the shared strategy and optimized transfer parameters. The transfer module layer includes multiple levels of transfer units, each corresponding to the hierarchical structure of the sparse model. The shared strategy and optimized transfer parameters can be loaded through the input layer and stored in different registers. Then, the knowledge transfer component architecture is adaptively optimized using the optimized transfer parameters. This adaptive optimization specifically includes: updating the parameters of each sub-unit of the transfer module layer layer by layer according to the learning rate in the optimized transfer parameters; using Adam for weight adjustment to ensure that each transfer unit has adaptive learning capabilities; and adjusting the weight decay coefficients in the transfer module layer according to the weight decay coefficients in the optimized transfer parameters. Dynamic adjustment of overfitting risk can suppress overtraining; dynamic pruning of low-contribution neurons according to the pruning ratio can preserve core feature pathways, reduce model complexity, and improve inference speed; finally, the optimized knowledge transfer component architecture is responded to layer by layer based on the sharing strategy, specifically including: monitoring the hierarchical response rate of each transfer unit through the rate calculation formula, and performing real-time calculation by combining historical data and real-time data; when the response rate of the sharing strategy reaches the preset index, the optimized knowledge transfer component architecture is used as the knowledge transfer component of the cloud collaborative architecture. The preset index can be set according to actual needs, and here it can be set to 90%, which is conducive to ensuring the reliability of the knowledge transfer component architecture and realizing intelligent sharing of clothing design resources and efficient scalability of the system in a multi-terminal environment.

[0042] It should be noted that the knowledge transfer component in this application refers to a core module in a cloud architecture used to realize multi-terminal clothing design resource sharing. It has functions such as model transfer, knowledge sharing, and adaptive generalization. Among them, the learning rate is used to control the step size of model training to prevent excessively fast convergence or oscillation; the weight decay is used to prevent model overfitting; the generalization ability of the model is improved by regularizing the weights; and the pruning ratio is used to dynamically prune low-contribution neurons, reduce model complexity, and improve inference speed. In addition, the layer-by-layer response in this embodiment refers to the knowledge transfer component architecture calculating the response rate layer by layer according to the policy requirements and performing dynamic optimization after loading the sharing policy, which is conducive to improving the collaborative efficiency of design tasks.

[0043] In step S5, the cloud-based collaborative architecture optimizes the collaborative process of clothing design resources based on the knowledge transfer component, generating shared clothing resources for intelligent generalization.

[0044] It should be noted that the collaborative process of apparel design resources refers to the integrated management of computing resources, knowledge base resources, and toolchain resources required for multiple design stages, such as product design, modeling, simulation, and optimization, through a unified data interface and scheduling mechanism. Based on task dependencies, it enables collaborative invocation across systems and modules, thereby supporting parallel design and dynamic interaction among multiple roles and processes.

[0045] In this embodiment, the cloud-based collaborative architecture optimizes the clothing design resources based on the knowledge transfer component to generate shared clothing resources for intelligent generalization. This can be achieved in the following way: Receive shared request information from the design terminal through a cloud-based collaborative architecture; The cloud-based collaborative architecture maps the shared request information hierarchically to multi-level migration units in the knowledge migration component based on the knowledge migration component, thereby obtaining hierarchical input feature vectors. In the knowledge transfer component, a multi-task transfer mechanism is used to optimize the shared transfer of hierarchical input feature vectors. When the shared adaptation rate is greater than the preset transfer performance index, the knowledge transfer component outputs clothing sharing resources for intelligent generalization.

[0046] In specific implementation, the design terminal initiates a sharing request to the cloud-based collaborative architecture through a preset service interface. The sharing request information includes resource type identification information, request strategy parameters, design terminal identification information, and historical resource call records. Specifically: the resource type identification information is used to indicate the category of clothing design resources to be requested; the request strategy parameters are used to limit the resource adaptation scope and optimization goals; the design terminal identification information includes the terminal device number, user identity credentials, and software environment information, which are used for permission verification and resource distribution configuration; the historical resource call records are used to enhance the migration and optimization capabilities of clothing shared resources; and the preset service interface can be built based on the RESTful API protocol.

[0047] In addition, upon receiving the sharing request information, the cloud-based collaborative architecture performs semantic analysis and field parsing on the content of the sharing request information using semantic analysis algorithms. The results of the semantic analysis and field parsing are then formatted to obtain a hierarchical input feature vector. It should be noted that semantic analysis and formatting can be performed using Python's natural language processing toolkit. Preferably, the spaCy library can be used for lexical analysis and named entity recognition of the sharing request information, combined with the pydantic library for field structure validation and format standardization. This helps the hierarchical input feature vector meet the JSON format requirements of the RESTful API interface specification.

[0048] In specific implementation, the knowledge transfer component employs a multi-task transfer mechanism to perform semantic transfer optimization on the hierarchical input feature vectors, obtaining the number of adaptable fields. The shared adaptability rate is calculated using the formula: Shared Adaptability Rate = Number of Adaptable Fields / Total Number of Fields. When the shared adaptability rate exceeds a preset transfer performance index, the knowledge transfer component outputs clothing sharing resources for intelligent generalization. These resources are stored in a cloud database and distributed and cached through the edge layer of the cloud collaborative architecture. The shared adaptability rate is an indicator of the ability of the target clothing design resources to be effectively adapted, reused, and generalized in multiple design terminals or diverse design task environments. The shared adaptability rate is a real value between 0 and 1. The preset transfer performance index can be set between 0.75 and 0.9 to help control the generalization degree of the clothing sharing resources; no specific limitation is made here.

[0049] It should be noted that the clothing shared resources in this application are design resources with unified structural expression, standardized semantic annotation, and support for heterogeneous design task invocation. In this application, the intelligent generalized clothing shared resources can facilitate the centralized management and intelligent distribution of clothing design resources in the cloud, thereby realizing the intelligent sharing of clothing design resources.

[0050] In summary, the technical solution adopted in this application can dynamically allocate and optimize the knowledge transfer of clothing design resources in a hierarchical manner, so as to realize the intelligent generalization and sharing of clothing design resources.

[0051] Example 2: This application provides an artificial intelligence-based clothing design system, referencing... Figure 4 As shown, this figure is a module structure diagram of an artificial intelligence-based clothing design system according to this application. The clothing design system includes: The multi-dimensional collaboration module 100 is used to determine the cloud collaboration architecture of clothing design resources, and then collect multi-dimensional design data through the cloud collaboration architecture; The layering and sparsification module 200 is used to dynamically layer and allocate the multidimensional design data to obtain the layered allocation result of the clothing design resources, and to perform sparsification processing on the multidimensional design data to obtain the sparsification model of the clothing design resources. The strategy training module 300 is used to acquire historical collaborative data of clothing design resources, and to train the sparse model based on deep reinforcement learning according to the hierarchical allocation results and the historical collaborative data, so as to obtain the sharing strategy and optimized transfer parameters of clothing design resources. Component building module 400 is used to build knowledge migration components of the cloud collaborative architecture based on the sharing strategy and the optimized migration parameters; The shared optimization module 500 is used by the cloud-based collaborative architecture to optimize the collaborative process of clothing design resources based on the knowledge transfer component, and generate clothing shared resources for intelligent generalization.

[0052] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0053] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0054] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A clothing design method based on artificial intelligence, characterized in that, The garment design method comprises the following steps: determine the cloud collaboration architecture of garment design resources, and then collect multi-dimensional design data through the cloud collaboration architecture; dynamically layering and distributing the multi-dimensional design data to obtain a layering and distribution result of the garment design resources, and sparsifying the multi-dimensional design data to obtain a sparsification model of the garment design resources; obtain historical collaboration data of the garment design resources, and train the sparsification model based on deep reinforcement learning according to the layering and distribution result and the historical collaboration data to obtain a sharing strategy and an optimized migration parameter of the garment design resources; construct a knowledge migration component of the cloud collaboration architecture based on the sharing strategy and the optimized migration parameter; the cloud collaboration architecture performs knowledge migration optimization on the collaboration process of the garment design resources based on the knowledge migration component to generate garment sharing resources for intelligent generalization.

2. The artificial intelligence-based garment design method of claim 1, wherein, The cloud collaboration architecture comprises a cloud, an edge layer and a terminal layer.

3. The artificial intelligence-based garment design method of claim 1, wherein, Collect multi-dimensional design data through the terminal layer in the cloud collaboration architecture, and the terminal layer comprises a plurality of design terminals, which are used to perform design tasks.

4. The artificial intelligence-based garment design method of claim 1, wherein, The dynamically layering and distributing the multi-dimensional design data to obtain a layering and distribution result of the garment design resources specifically comprises: determine a multi-dimensional index set of the multi-dimensional design data; dynamically clustering the multi-dimensional design data based on the multi-dimensional index set to obtain different clustering clusters; layering all the clustering clusters to obtain a layering and distribution result of the garment design resources.

5. The artificial intelligence-based garment design method of claim 1, wherein, The sparsifying the multi-dimensional design data to obtain a sparsification model of the garment design resources specifically comprises: performing an unstructured pruning operation on the preprocessed multi-dimensional design data to obtain pruned data; determining a sparsity ratio of the garment design resources based on a preset mask matrix and the pruned data; adapting the sparsity ratio to obtain a sparsification model of the garment design resources.

6. The artificial intelligence-based garment design method of claim 1, wherein, Obtain the historical collaboration data of the garment design resources through the operation logs of the design terminals in the cloud collaboration architecture.

7. The artificial intelligence-based garment design method of claim 1, wherein, Train the sparsification model based on deep reinforcement learning according to the layering and distribution result and the historical collaboration data to obtain a sharing strategy and an optimized migration parameter of the garment design resources specifically comprises: construct a layering reward function through the layering and distribution result; train the sparsification model based on the historical collaboration data through the layering reward function to obtain an optimized migration parameter of the garment design resources; use a deep deterministic policy gradient algorithm to perform policy estimation on the sparsification model based on the historical collaboration data to obtain a sharing strategy of the garment design resources.

8. The artificial intelligence-based garment design method of claim 1, wherein, Constructing a knowledge migration component of the cloud collaboration architecture based on the sharing strategy and the optimized migration parameter specifically comprises: initializing a knowledge migration component architecture, and loading the sharing strategy and the optimized migration parameter through the knowledge migration component architecture; the optimized migration parameter adaptively optimizes the generalization ability of the knowledge migration component architecture according to a learning rate, a weight decay and a pruning ratio to obtain an optimized knowledge migration component architecture; The optimized knowledge migration component architecture is responded to layer by layer based on the sharing strategy, and when the response rate of the sharing strategy reaches a preset index, the optimized knowledge migration component architecture is taken as a knowledge migration component of the cloud collaboration architecture.

9. The artificial intelligence-based garment design method of claim 1, wherein, The knowledge migration component is a multi-level transfer learning unit based on federated transfer learning.

10. An artificial intelligence-based garment design system for performing an artificial intelligence-based garment design method according to any one of claims 1 to 9, characterized by, The garment design system comprises: a multi-dimensional collaboration module configured to determine a cloud collaboration architecture of garment design resources, and to collect multi-dimensional design data through the cloud collaboration architecture; a hierarchical and sparse module configured to dynamically allocate the multi-dimensional design data in layers to obtain a hierarchical allocation result of the garment design resources, and to perform sparse processing on the multi-dimensional design data to obtain a sparse model of the garment design resources; a strategy training module configured to obtain historical collaboration data of the garment design resources, to perform strategy training on the sparse model based on deep reinforcement learning according to the hierarchical allocation result and the historical collaboration data, and to obtain a sharing strategy and optimized migration parameters of the garment design resources; a component construction module configured to construct a knowledge migration component of the cloud collaboration architecture based on the sharing strategy and the optimized migration parameters; a sharing optimization module configured to perform knowledge migration optimization on a collaboration process of the garment design resources based on the knowledge migration component of the cloud collaboration architecture, and to generate garment sharing resources for intelligent generalization.

Images