AIGC-based traditional pattern generation and industry adaptation method and system

Abstract

The invention belongs to the technical field of traditional pattern design, and provides a traditional pattern generation and industrial adaptation method and system based on AIGC, and the technical scheme is that the method comprises the steps: obtaining traditional pattern image data, and carrying out the preprocessing of the traditional pattern image data; generating a traditional pattern image based on the preprocessed traditional pattern image data and the trained pattern generation model; determining to-be-modified local area information in the traditional pattern image, and modifying the to-be-modified local area information to obtain the traditional pattern image after the local area is modified; and matching the generated pattern image with the industrial product to obtain an industrial product matching scheme. And digital transformation and efficient industrialization of the traditional pattern design and the cultural creative industry are promoted.

Description

Technical Field

[0001] This invention belongs to the field of traditional pattern design technology, and in particular relates to a method and system for generating and adapting traditional patterns to industries based on AIGC. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Traditional patterns, as an important component of Chinese visual culture, are widely used in various fields such as architectural decoration, clothing design, and cultural and creative products, carrying rich historical connotations and aesthetic value. Currently, the application of traditional patterns in modern design faces numerous challenges. Existing pattern design processes have low automation levels and long design cycles, making it difficult to meet the market demands for rapid iteration and mass production. Although Artificial Intelligence Generated Content (AIGC) technology has made significant progress in image generation, the resolution, modifiability, and cultural semantic expression of AIGC-generated patterns still fail to meet industry requirements, resulting in pattern designs that cannot be adapted to industry needs. Furthermore, when adapting to industry, there is a lack of unified and reproducible quantitative evaluation standards for assessing the quality of existing pattern designs. In addition, applying generated patterns to specific products requires extensive manual adaptation and adjustment work, such as layout, color matching, resolution conversion, and texture simulation for different materials, further increasing the complexity and cost of design and production, hindering the rapid industrialization of pattern design results. Summary of the Invention

[0004] To address at least one of the technical problems mentioned above, this invention provides a method and system for generating and adapting traditional patterns to industries based on AIGC, which significantly improves the industrial applicability of generated patterns and promotes the digital transformation and efficient industrialization of traditional pattern design and cultural and creative industries.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a traditional pattern generation and industry adaptation method based on AIGC, comprising the following steps: Acquire traditional pattern image data and preprocess the traditional pattern image data; Traditional pattern images are generated based on preprocessed traditional pattern image data and a trained pattern generation model; the construction process of the pattern generation model includes: A training dataset is constructed based on preprocessed traditional pattern image data. A noise prediction network is trained on the constructed training dataset to obtain a trained noise prediction network. The input control conditions are determined, and the trained noise prediction network is controlled according to the control conditions to generate traditional pattern images that meet the control conditions. A structured request is constructed, and a traditional pattern image optimization strategy is generated based on the structured request and a large language model. The traditional pattern images that meet the control conditions are optimized based on the generated traditional pattern image optimization strategy to obtain optimized traditional pattern images. The local regions to be modified in the optimized traditional pattern image are identified, and the information of the local regions to be modified is modified to obtain the final traditional pattern image. The final traditional pattern image is adapted to industrial products to obtain an industrial product adaptation solution.

[0006] Furthermore, the preprocessing of traditional pattern image data includes standardization, multi-dimensional annotation, and knowledge graph construction.

[0007] Further, the training of the noise prediction network based on the constructed training dataset to obtain the trained noise prediction network includes: pattern image As input, it is passed through a variational autoencoder encoder. The latent space is compressed to a lower dimension to obtain the latent representation. ; This potential representation Simulate a forward diffusion process, in T Gaussian noise is gradually added to it within each time step to obtain the noisy latent representation. ; Text description Unstructured text condition vectors are obtained through text encoder encoding. The subgraphs related to the current text prompt are retrieved from the knowledge graph and encoded into structured semantic vectors. ; Noisy latent representation z t Unstructured text condition vectors and structured semantic vectors The input is fed into the U-Net network for training to minimize a loss function based on mean squared error. The trained noise prediction network is obtained as the optimization objective.

[0008] Further, determining the input control conditions and controlling the trained noise prediction network according to the control conditions to generate a traditional pattern image that meets the control conditions includes: When the control condition is textual, the input prompt word is converted into a text embedding vector through a cross-modal encoder. This text embedding vector is then used as a conditional input to the U-Net network of the diffusion model to guide the denoising process. When the control condition is style / image condition control, the U-Net network parameters are completely frozen, and a parameter-trainable copy is created as a ControlNet branch. In the ControlNet branch, the pattern template or reference image is preprocessed into a conditional control map. The conditional control map is input into the trainable branch of ControlNet, and the extracted features are injected into the generation process of the locked branch U-Net to generate a traditional pattern image that meets the control conditions.

[0009] Furthermore, the optimization strategy for generating traditional pattern images based on structured requests and large language models includes: After the large language model receives a structured request, it addresses unsatisfactory images. I bad Perform internal visual analysis and convert it into structured visual metadata. M vision To provide users with natural language feedback F user and M vision and original prompt words Po rig Semantic association is performed, and traditional pattern image optimization strategies are generated based on semantic association information, namely, a new structured cue word. P new .

[0010] Furthermore, the method also includes multi-dimensional quantitative evaluation of the generated traditional pattern images, including evaluation of the traditional dimension, aesthetic dimension and innovative dimension, respectively, to obtain the traditional dimension score, aesthetic score and innovative dimension score, and the comprehensive evaluation result is obtained by weighting the scores of each dimension.

[0011] Furthermore, when adapting the generated pattern images to industrial products, this includes layout adaptation, color adaptation, format conversion, and process parameter adaptation.

[0012] A second aspect of the present invention provides a traditional pattern generation and industry adaptation system based on AIGC, comprising: The image preprocessing module is used to acquire traditional pattern image data and preprocess the traditional pattern image data. The pattern generation module is used to generate traditional pattern images based on preprocessed traditional pattern image data and a trained pattern generation model; the construction process of the pattern generation model includes: A training dataset is constructed based on preprocessed traditional pattern image data. A noise prediction network is trained on the constructed training dataset to obtain a trained noise prediction network. The input control conditions are determined, and the trained noise prediction network is controlled according to the control conditions to generate traditional pattern images that meet the control conditions. A structured request is constructed, and a traditional pattern image optimization strategy is generated based on the structured request and a large language model. The traditional pattern images that meet the control conditions are optimized based on the generated traditional pattern image optimization strategy to obtain optimized traditional pattern images. The pattern modification module is used to determine the local area information to be modified in the optimized traditional pattern image, modify the local area information to be modified, and obtain the final traditional pattern image. The industry adaptation module is used to adapt the final traditional pattern image to industrial products to obtain an industrial product adaptation solution.

[0013] A third aspect of the present invention provides a computer-readable storage medium.

[0014] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a conventional pattern generation and industry adaptation method based on AIGC as described above.

[0015] A fourth aspect of the present invention provides a computer device.

[0016] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a conventional pattern generation and industry adaptation method based on AIGC as described above.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention optimizes traditional pattern images step by step by setting control conditions and constructing structured requests. By integrating high-resolution reconstruction technology and local modification technology, it ensures that the generated patterns have the clarity and precision required for industrial-grade printing, and allows users to make precise adjustments to local elements, thus achieving fine control over the generated results.

[0018] This invention significantly shortens the design cycle through automated and intelligent pattern generation, evaluation, and adaptation processes. The complex process of traditional pattern design, from creative conception to finished product application, can be reduced from weeks or months to minutes or even seconds, enabling rapid response and mass production, and greatly improving design efficiency.

[0019] This invention, by constructing a multi-dimensional knowledge graph of traditional patterns and employing a conditional diffusion model, achieves cross-disciplinary integration and re-creation of pattern elements from different periods, styles, and themes. This not only efficiently reproduces and protects traditional culture but also enables innovation based on digitalization, providing a completely new path for the modern application of traditional patterns.

[0020] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0021] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0022] Figure 1 This is a flowchart of a traditional pattern generation and industry adaptation method based on AIGC provided in an embodiment of the present invention; Figure 2 This is a block diagram of a traditional pattern generation and industry adaptation system based on AIGC provided in an embodiment of the present invention. Detailed Implementation

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0024] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0025] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0026] Despite significant progress in image generation technology, Artificial Intelligence Generated Content (AIGC) still presents the following challenges when applied to traditional pattern design generation: 1. Resolution and Accuracy Limitations. Images generated by existing AIGC models often perform poorly in terms of resolution and detail accuracy, making it difficult to meet the precision and clarity requirements of traditional patterns on products (such as textiles and printed materials). This leads to a significant need for manual correction of the generated patterns in practical applications, reducing the level of automation.

[0027] 2. Irregularity of Generated Images. Most AIGC models offer limited user capabilities for fine-tuning local elements, structures, or details after generating images. This means designers cannot directly fine-tune AI-generated images to meet specific needs, still requiring time-consuming and labor-intensive modifications using traditional design tools (such as Adobe Illustrator and Adobe Photoshop), hindering the deep integration of AIGC into professional design workflows.

[0028] 3. Lack of cultural semantic expression. Existing AIGC models are insufficient in understanding and accurately representing complex and abstract "traditional cultural semantics". When users describe traditional patterns through prompts, the model may struggle to accurately capture their cultural connotations, historical styles, symbolic meanings, or the precise forms of specific pattern elements, leading to deviations between the generated results and user expectations, or even cultural misinterpretations or stylistic confusion.

[0029] 4. Stylistic rigidity and innovation bottleneck. Existing pattern designs are often limited to the style of specific designers or historical templates, lacking novelty and the possibility of cross-cultural integration.

[0030] 5. Difficulty in adapting to industry. When applying the generated patterns to specific products, a large amount of manual adaptation and adjustment work is required, such as layout, color matching, resolution conversion, and texture simulation for different materials. This further increases the complexity and cost of design and production, hindering the rapid industrialization of pattern design results.

[0031] To address the problems of low resolution, unmodifiable nature, inaccurate cultural semantic expression, and difficulty in industry adaptation in AIGC-generated patterns, this application proposes a traditional pattern generation and industry adaptation method based on AIGC. Through automated and intelligent pattern generation, evaluation, and adaptation processes, the design cycle is greatly shortened. The complex process of traditional pattern design from creative conception to finished product application can be reduced from weeks or months to minutes or even seconds, achieving rapid response and mass production, and greatly improving design efficiency.

[0032] Example 1 like Figure 1 As shown, this embodiment provides a traditional pattern generation and industry adaptation method based on AIGC, including the following steps: Step 1: Obtain traditional pattern image data and preprocess the traditional pattern image data; In this embodiment, when acquiring traditional pattern image data, web crawling technology, image scanning technology, and manual input are used to systematically collect traditional pattern images from publicly available sources such as digital collections of museums, historical documents, art albums, and high-definition photos of cultural relics.

[0033] In this embodiment, the preprocessing of traditional pattern image data includes standardization, multi-dimensional annotation, and knowledge graph construction; This process involves standardizing traditional pattern image data, such as unifying image size, color correction, background removal, and noise reduction. Furthermore, data augmentation techniques like rotation, flipping, cropping, and color shifting are used to expand the dataset and improve the model's generalization ability. Additionally, vectorization tools (such as Potrace) are employed to enhance image clarity and convert it to an industrially applicable vector graphics format. Multi-dimensional annotation includes a combination of manual annotation and AI-assisted annotation, such as basic pattern information: pattern type, historical period; semantic tags: pattern theme, element composition, symbolic meaning; style tags: artistic style, color system; structural tags: composition type, line characteristics.

[0034] The construction of the knowledge graph involves associating labeled multi-dimensional tags and pattern image data, and using graph database technologies (such as Neo4j) or knowledge graph building tools to store these entities and relationships as triples, thereby constructing a traditional pattern knowledge graph. This graph can clearly represent the relationships between different patterns, periods, cultures, and styles, providing a structured and queryable knowledge base for traditional evaluation in the pattern generation module. The structured data of this knowledge graph will serve as the core of semantic understanding and will be deeply integrated into subsequent pattern generation to ensure the cultural accuracy of the generated content and the interpretability of the evaluation results.

[0035] Step 2: Generate traditional pattern images based on the preprocessed traditional pattern image data and the trained pattern generation model; In this embodiment of the invention, the pattern generation model is technically a Conditional Latent Diffusion Model (LDM). The core component of this model is a noise prediction network for predicting noise, which in this embodiment adopts the U-Net architecture. Therefore, in the following text, the noise prediction network specifically refers to the implementation of this U-Net architecture. The specific steps include: Step 201: Construct a training dataset based on the preprocessed traditional pattern image data, and train the noise prediction network based on the constructed training dataset to obtain the trained noise prediction network. Step 2011: Construct a training dataset based on the preprocessed traditional pattern image data; In constructing the training dataset based on preprocessed traditional pattern image data, the dataset consists of paired data, with each pair containing a high-resolution pattern image from the constructed traditional pattern dataset, denoted as . x And the corresponding natural language text description sentence constructed from multi-dimensional metadata tags, denoted as c ; Step 2012: Train the noise prediction network based on the constructed training dataset to obtain the trained noise prediction network, including: Step 20121: Input pattern image The encoder is obtained through a variational autoencoder (VAE). The latent space compressed to a lower dimension yields the latent representation. ; Step 20122, regarding the potential representation Simulate a forward diffusion process, in T Gaussian noise is gradually added to it within each time step. Obtain the latent representation after adding noise z t ; The calculation formula is: , in, It is standard Gaussian noise. α t It is based on a predefined noise schedule at time steps. t The coefficients are defined, for example, using a cosine schedule or a linear schedule. They are not variables that need to be obtained in real time, but are set as hyperparameters. Step 20123: Describe the text Unstructured text conditional vectors are obtained through text encoders (such as CLIP, Contrastive Language-Image Pre-training). The subgraphs related to the current text prompt are retrieved from the knowledge graph and encoded into structured semantic vectors. ; Step 20124: Add noise to the latent representation z t Unstructured text condition vectors and structured semantic vectors The input is fed into the U-Net network to predict the added noise, and this network is denoted as the original, fine-tuned Stable Diffusion model. In this embodiment, to address the technical problem that existing U-Net networks can only understand unstructured text prompts and cannot accurately express the structured cultural semantics of traditional patterns, the attention mechanism of U-Net has been improved, specifically including: Traditionally, U-Net uses cross-attention layers in its middle blocks and upsampling blocks to receive text conditions. The middle blocks and upsampling blocks in this application employ a gated fusion attention mechanism to receive structured semantic vectors in parallel. Unstructured text condition vectors The latent representation after adding noise z t Q, as a gated fusion attention mechanism, will use structured semantic vectors Unstructured text condition vectors As Key and Value, structured semantic vectors are dynamically and adaptively fused through a trainable gating parameter G (a scalar or vector between 0 and 1). Unstructured text condition vectors Obtain the fused features , is represented as: , Furthermore, the parameters of the noise prediction network are optimized using the backpropagation algorithm, with the goal of minimizing a loss function based on the mean squared error (MSE). L This means minimizing the difference between the predicted noise and the actual noise, and its formula is: , in, Indicates in data x ,condition c ,noise ε and time step t Expected Value This indicates that at time step t Added real Gaussian noise, ε θThis indicates that the noise prediction network is based on U-Net (i.e., the model parameters are...). θ The predicted noise; τ θ ( c ) is a text conditional vector generated by a text encoder (such as CLIP).

[0036] Step 202: Determine the input control conditions, and control the trained noise prediction network according to the control conditions to generate a traditional pattern image that meets the control conditions; In this embodiment, the input control conditions include two mechanisms: text condition control mechanism and style / image condition control; When the control condition is a text-based conditional control mechanism, the prompts input by the user are converted into text embeddings through cross-modal encoders such as CLIP. These embedding vectors are then used as conditional inputs into the U-Net network of the diffusion model to guide the denoising process and ensure that the final generated image conforms to the text description.

[0037] When the control condition is style / image condition control, a two-branch structure is adopted when selecting a pattern template or uploading a reference image. The first branch completely freezes the U-Net network parameters of the original, fine-tuned Stable Diffusion model. The other branch uses ControlNet as an additional condition control module to add and fuse the structural features output by ControlNet at each level with the features of the encoder layer and intermediate blocks corresponding to the locked branch U-Net network to obtain the final feature vector that conforms to style / image condition control. Specifically, when a user provides a template image as a reference, the image is first preprocessed into a conditional map. For example, the contour map is extracted using the Canny edge detection algorithm, which carries precise structural information about the control chart. The feature vectors extracted from the trainable branches of ControlNet are then fed into the generation process of the locked branches; this process can be represented as: , in, It is the locked branch decoder. i The output features of the layer It is a new feature after the injection of control information. F It is a ControlNet network, and its parameters are: , Z It is a zero convolutional layer with the following parameters: , It is the input conditional control graph. Since the weights of the zero convolution are initialized to zero at the beginning of training, the stability of training is guaranteed, so that structural information can be smoothly incorporated.

[0038] This mechanism allows text prompts to primarily control the content and style of the generated pattern, while the conditional control graphs of the ControlNet input (such as the outline of the template) precisely dominate the composition, structure, and line layout of the final image, thus ensuring that the generated pattern is highly consistent with the template provided by the user in terms of composition.

[0039] Step 203: Construct a structured request, generate a traditional pattern image optimization strategy based on the structured request and the large language model, optimize the traditional pattern image that meets the control conditions based on the generated traditional pattern image optimization strategy, and obtain the optimized traditional pattern image. To address the issue of unsatisfactory generated images, a closed-loop prompt word optimization function based on a Large Language Model (LLM) is provided. The specific implementation steps are as follows: Step 2031: Construct a structured request; When a user submits feedback (such as "the lines are not smooth enough" or "the colors are too dark"), the system does not directly forward the text. Instead, it constructs a structured request, presenting the original prompt text. Po rig User natural language feedback F user The generated image "Ibad" does not meet expectations, and a structured request is sent to the API of the Big Prophecy model; among which, the original prompt words Po rig Based on the text descriptions in the training dataset, images that do not meet the expected results (Ibad) can be manually filtered or filtered using existing algorithms. Step 2032: Optimize traditional pattern image generation strategies based on structured requests and large language models; After receiving a structured request, the large language model first processes unsatisfactory images. I bad Perform internal visual analysis and convert it into structured visual metadata. M vision Then, the user's natural language feedback F user and M vision and original prompt words Po rig Semantic association is performed, and then a traditional pattern image optimization strategy is generated based on the semantic association information, specifically a new structured cue word. P new ; For example,M vision = { elements : ['Dragon', 'Cloud'], style : 'Tang Dynasty' composition :'symmetry', color analysis : { brightness 0.3, contrast 0.4. LLM will use the user's natural language feedback. F user (For example, "the lines are not smooth enough") and M vision For example,{ elements : ['Dragon']}) and P orig (For example, "Tang Dynasty dragon patterns") are semantically associated; LLM generates optimization strategies based on this semantic association, for example, if F user If the line is not smooth enough, LLM will determine whether to add descriptive words to control the quality of the line (such as "vector art" or "clear outline") or add negative prompts (such as "blurry" or "sketch").

[0040] Step 2033: Optimize the traditional pattern image that meets the control conditions based on the generated traditional pattern image optimization strategy to obtain the optimized traditional pattern image; Based on the generated optimization strategy, LLM rewrites or enhances the prompt word while retaining the original core elements, automatically generating a new, more accurate, and descriptive prompt word. The system then uses this optimized new prompt word. P new The text is automatically re-input into the semantic encoder (such as CLIP) of the diffusion model to generate new text conditional vectors. τ θ ( c new This process drives a new round of noise reduction and generation. This human-computer interactive closed-loop optimization process does not require users to have professional knowledge of prompt word engineering. It can guide the generation model to gradually approach its final design goal through natural language feedback, thereby more accurately meeting user needs.

[0041] Step 3: Determine the local area information to be modified in the optimized traditional pattern image, modify the local area information to obtain the traditional pattern image after local area modification; The high-resolution generation submodule can enlarge the pattern after it is generated using AI super-resolution technology, such as ESRGAN (Enhanced Super-Resolution Generative Adversarial Networks) or SwinIR models. While enlarging the image, it fills in lost details, ensuring the pattern achieves the clarity and resolution required for industrial-grade printing. The local modification and iterative optimization submodule allows operation via an image mask-based interactive interface. Users can select the areas of the pattern that need modification and use prompts or parameters, such as "change this pattern to a cloud pattern" or "adjust to gold," to drive a local inpainting model. This modifies only specific areas while preserving the integrity of other parts. This fine-grained control greatly improves the controllability and usability of pattern generation, meeting personalized needs. Specifically, it includes the following steps: Step 301: Receive the information of the local area to be modified in the traditional pattern image, and process the information of the local area to be modified; In this embodiment, the system receives local region information selected by the user through the interactive interface and converts it into a binary mask tensor with the same size as the original image. The pixel values ​​of the region to be modified are 1, and the values ​​of the reserved region are 0. Simultaneously, the system receives new prompts or parameters input by the user as text conditions.

[0042] Step 302: Generate new control conditions; The new text conditions are converted into a high-dimensional conditional embedding vector by a text encoder (such as CLIP) to guide the semantics of the generated content. Next, the system fills the areas to be modified in the original image tensor, specified by the mask tensor, with random noise, while leaving the pixel data of the reserved areas unchanged, forming a noisy conditional image tensor.

[0043] Step 303: Guide local redrawing generation based on new control conditions; The noisy image tensor, mask tensor, and conditional embedding vector are fed together into the local inpainting model. During the iterative denoising process, the model is precisely guided by two conditions. Specifically, this includes the following steps: at each time step of the denoising process... t The model is first based on z t and new prompt words c new Predict the denoised latent representation z 0_pred Then, a forced replacement is performed using the mask tensor M: The latent representation of the original image (unmodified region) z 0_orig Re-add noise to the current time step t ,get z t_orig ; The newly generated region (within the mask) M =1) and the original noisy region (outside the mask) M =0) Synthesis is performed, and the result of the previous synthesis step is used as the input for the next denoising step. By resetting the area outside the mask to the noisy representation of the original image in each step, the pattern image result of local reversion is obtained, which forces the content generated by the model to achieve a seamless and natural connection with the boundary of the preserved area.

[0044] Step 304: Based on the results of local redrawing and the original traditional pattern image, synthesize to obtain the final traditional pattern image; Through this series of automated data processing steps, the system of this invention enables precise and controllable modification of specific areas of a pattern while fully preserving the rest.

[0045] This system addresses the problems of low resolution, lack of modifiability, and inaccurate cultural semantic expression in existing AIGC-generated patterns. By integrating high-resolution reconstruction technology and a local modification submodule, it ensures that the generated patterns possess the clarity and precision required for industrial-grade printing and allows users to precisely adjust local elements, achieving refined control over the generated results.

[0046] Step 4: Perform multi-dimensional quantitative evaluation on the generated traditional pattern images to obtain a comprehensive evaluation result; In this embodiment, the multi-dimensional evaluation specifically includes evaluation based on traditional, aesthetic, and innovative dimensions. Specifically, the steps include the following: Step 401: Calculate the traditionality dimension score for the generated traditional pattern image in the traditionality dimension. ; In this embodiment, a weighted fusion algorithm with CNN-KG (Knowledge Graph) dual verification is used to adjust the final score representation as follows: , in, The confidence score of a CNN model (such as ResNet) for the overall style. S cnn The proportion of elements that passed verification out of the total number of identified elements. The knowledge graph validation score is calculated by using an object detection model (such as YOLOv8 or DETR) to identify the core design elements in the generated pattern.E = { e 1, e 2,...}, and then, query these design elements in the knowledge graph. e i and S cnn The relevance of the target style to be determined. and These are the weighting coefficients; Step 402: Calculate the aesthetic score of the generated traditional pattern image in the aesthetic dimension; In this embodiment, a pre-trained model based on the NIMA (Neural Image Assessment) architecture is explicitly used. This model is trained on an aesthetic rating dataset (such as the AVA Dataset), and its output is not a single score, but a probability distribution of scores from 1 to 10. This application uses the expected value of this distribution as the final quantified aesthetic score. Step 403: Calculate the innovation dimension score for the generated traditional pattern image; The innovativeness dimension score includes two dimensions: novelty and cross-disciplinaryness. Novelty is calculated by comparing the feature vector of the new pattern with the feature vector of all patterns in the database. Cross-boundary score is calculated by measuring the graph distance of the semantic elements fused in the generated pattern within the knowledge graph. The score for the innovation dimension is obtained by weighting the scores from both novelty and cross-boundary score dimensions. Step 404: The final comprehensive evaluation result is obtained by weighting the scores from each dimension.

[0047] This invention establishes a multi-dimensional quantitative evaluation system, transforming abstract design concepts such as tradition, aesthetics, and innovation into calculable numerical indicators. This provides an objective and scientific basis for communication between designers and clients, reducing misunderstandings and repeated revisions caused by subjective judgments, thereby significantly reducing project costs and time.

[0048] Step 5: Adapt the generated pattern image to the industrial product to obtain the adaptation solution; Specifically, the steps include the following: Step 501: Adapt the generated pattern image to the industrial product layout; Intelligently apply two-dimensional pattern images to two-dimensional or three-dimensional product models; For two-dimensional product models (such as clothing pattern diagrams), the system employs deep learning-based semantic segmentation technology to automatically identify key regions. Specifically, the system incorporates a semantic segmentation model (such as U-Net) pre-trained on a product pattern diagram dataset. Upon receiving a product model image, the model outputs a pixel-level segmentation mask that accurately identifies applicable "key regions" such as "front panel" and "sleeves." Subsequently, based on the user-selected region, the pattern image is scaled, cropped, and filled into the outline of that region.

[0049] For 3D product models (such as bottles, jars, and cushions), the UV mapping data is obtained by parsing the 3D model file (such as .obj or .fbx format). This UV data provides a UV map showing the unfolded surface of the 3D model onto a 2D plane. The system applies (textures) the pattern image onto this 2D UV map. Subsequently, the rendering engine uses the model's built-in UV coordinates to reverse-map the image pixels on the UV map back to the corresponding surface of the 3D model, and applies an image warping algorithm to ensure that the pattern can be correctly stretched and wrapped according to the curvature of the model, achieving a seamless fit.

[0050] Step 502: Match the generated pattern image with the colors of the industrial product; The system automatically adjusts the color scheme of the pattern based on the material, brand color, or user-specified color system of the target product. It receives the source pattern image and the target color system as input and processes them using the following appropriate technical approaches, depending on the type of the target system: When the target is a specified color system (such as a brand color chart), the system uses a cluster-based color quantization method, which includes the following steps: Convert both the source image and the target color chart (such as a brand color chart) to... CIELAB Color space, because the Euclidean distance in this space is more consistent with the color difference perceived by humans; K-Means clustering was performed on the Lab pixel data of the source image to obtain K cluster centers. C src = { c s1 ,..., c sk}, for each source cluster center c si Using Euclidean distance to the target color chart P tgt = { c t1 ,... c tm Find the color that is closest in color perception in the color} and calculate the minimum Euclidean distance.

[0051] When the target is a reference image, the system employs a color transfer algorithm based on color space statistical features, specifically including: First, both the source and reference images are converted to the Lab color space, which can separate luminance and chromaticity information; Next, the mean values ​​of the two images in the L, a, and b channels are calculated respectively. μ src , μ ref ) and standard deviation ( σ src , σ ref ); Then perform a mathematical transformation: iterate through each pixel Psrc of the source image and apply the following transformation formula to adjust the values ​​of its L, a, and b channels: P new_channel = ( ( P src_channel - μ src ) ·( σ ref / σ src ) ) + μ ref , in, P new_channel This represents the pixel value after channel adjustment. P src_channel This represents the pixel value before channel adjustment. μ src and μ ref This represents the mean values ​​of the source and reference images across the L, a, and b channels. σ src , σ reff This represents the standard deviation of the source and reference images across the L, a, and b channels. Finally, the pixel values ​​after channel adjustment are converted back to RGB space to achieve the overall tone and style transfer.

[0052] When the target is a Look-Up Table (LUT), the system performs direct color mapping. The LUT file defines the precise correspondence between input and output colors. The system iterates through each pixel of the source image, using its RGB value as an index, and searches for and replaces it with the corresponding output color value in the LUT to efficiently achieve specific filter or stylization effects.

[0053] Step 503: Convert the format of the generated pattern image and the industrial product and adapt the process parameters; In this embodiment, it is necessary to adapt according to the set rules based on the process parameters; For example, when the process parameter is screen printing, enter the number of printing colors. N (For example N =6), perform color quantization (e.g., using K-Means or Octree algorithms) to reduce the image colors to 6. N Color, generation N Create separate color layers and export them as .PSD or .AI format; When the process parameter is embroidery, enter the minimum stitch width (Min Stitch Width). W min ), number of embroidery thread colors M (For example M =12), quantize the color to M Color, identify all fine lines in the image, and use morphological dilation operations (e.g.) cv 2. dilate Ensure that the width of all lines is greater than [the specified value]. W min， Convert to a vector path and export as a .DST or .EMB file.

[0054] Through this step, the system's product adaptation module supports diverse product application needs, such as clothing, home textiles, packaging and printing, home decoration, and cultural and creative products. It provides automatic matching of multiple parameters and export of industrial standard formats, which greatly improves the industrial applicability of generated patterns and promotes the digital transformation and efficient industrialization of traditional pattern design and cultural and creative industries.

[0055] Example 2 like Figure 2 As shown, this embodiment provides a traditional pattern generation and industry adaptation system based on AIGC, including: The image preprocessing module is used to acquire traditional pattern image data and preprocess the traditional pattern image data. The pattern generation module is used to generate traditional pattern images based on preprocessed traditional pattern image data and a trained pattern generation model; the construction process of the pattern generation model includes: A training dataset is constructed based on preprocessed traditional pattern image data. The noise prediction network is trained on the constructed training dataset to obtain the trained noise prediction network. The input control conditions are determined, and the trained noise prediction network is controlled according to the control conditions to generate traditional pattern images that meet the control conditions. A structured request is constructed, and an optimization strategy for traditional pattern images is generated based on the structured request and a large language model. The pattern modification module is used to determine the local area information to be modified in the traditional pattern image, modify the local area information to obtain the traditional pattern image after local area modification; The industry adaptation module is used to adapt the generated pattern images to industrial products to obtain industrial product adaptation solutions.

[0056] It should be noted that the specific implementation of the AIGC-based traditional pattern generation and industry adaptation system in this embodiment of the invention is similar to the specific implementation of the AIGC-based traditional pattern generation and industry adaptation method in this embodiment of the invention. Please refer to the description in the method section for details. To reduce redundancy, it will not be repeated here.

[0057] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the traditional pattern generation and industry adaptation method based on AIGC described above.

[0058] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the traditional pattern generation and industry adaptation method based on AIGC described above.

[0059] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0060] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. 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 flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0061] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0062] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0063] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0064] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A traditional pattern generation and industry adaptation method based on AIGC, characterized in that, Includes the following steps: Acquire traditional pattern image data and preprocess the traditional pattern image data; Traditional pattern images are generated based on preprocessed traditional pattern image data and a trained pattern generation model; the construction process of the pattern generation model includes: A training dataset is constructed based on preprocessed traditional pattern image data. A noise prediction network is trained on the constructed training dataset to obtain a trained noise prediction network. The input control conditions are determined, and the trained noise prediction network is controlled according to the control conditions to generate traditional pattern images that meet the control conditions. A structured request is constructed, and a traditional pattern image optimization strategy is generated based on the structured request and a large language model. The traditional pattern images that meet the control conditions are optimized based on the generated traditional pattern image optimization strategy to obtain optimized traditional pattern images. The local regions to be modified in the optimized traditional pattern image are identified, and the information of the local regions to be modified is modified to obtain the final traditional pattern image. The final traditional pattern image is adapted to industrial products to obtain an industrial product adaptation solution.

2. The method for generating traditional patterns and adapting them to industry based on AIGC as described in claim 1, characterized in that, The preprocessing of traditional pattern image data includes standardization, multi-dimensional annotation, and knowledge graph construction.

3. The method for generating traditional patterns and adapting them to industry based on AIGC as described in claim 1, characterized in that, The training of the noise prediction network based on the constructed training dataset to obtain the trained noise prediction network includes: pattern image As input, it is passed through a variational autoencoder encoder. The latent space is compressed to a lower dimension to obtain the latent representation. ; This potential representation Simulate a forward diffusion process, in T Gaussian noise is gradually added to it within each time step to obtain the noisy latent representation. ; Text description Unstructured text condition vectors are obtained through text encoder encoding. The subgraphs related to the current text prompt are retrieved from the knowledge graph and encoded into structured semantic vectors. ; Noisy latent representation z t Unstructured text condition vectors and structured semantic vectors The input is fed into the U-Net network for training to minimize a loss function based on mean squared error. The trained noise prediction network is obtained as the optimization objective.

4. The method for generating traditional patterns and adapting them to industry based on AIGC as described in claim 1, characterized in that, The determination of input control conditions, and the control of the trained noise prediction network based on the control conditions to generate a traditional pattern image that meets the control conditions, includes: When the control condition is textual, the input prompt word is converted into a text embedding vector through a cross-modal encoder. This text embedding vector is then used as a conditional input to the U-Net network of the diffusion model to guide the denoising process. When the control condition is style / image condition control, the U-Net network parameters are completely frozen, and a parameter-trainable copy is created as a ControlNet branch. In the ControlNet branch, the pattern template or reference image is preprocessed into a conditional control map. The conditional control map is input into the trainable branch of ControlNet, and the extracted features are injected into the generation process of the locked branch U-Net to generate a traditional pattern image that meets the control conditions.

5. The method for generating traditional patterns and adapting them to industry based on AIGC as described in claim 1, characterized in that, The optimization strategy for generating traditional pattern images based on structured requests and large language models includes: After the large language model receives a structured request, it addresses unsatisfactory images. I bad Perform internal visual analysis and convert it into structured visual metadata. M vision To provide users with natural language feedback F user and M vision and original prompt words Po rig Semantic association is performed, and traditional pattern image optimization strategies are generated based on semantic association information, namely, a new structured cue word. P new .

6. The method for generating traditional patterns and adapting them to industry based on AIGC as described in claim 1, characterized in that, The method also includes multi-dimensional quantitative evaluation of the generated traditional pattern images, including evaluation of the traditional dimension, aesthetic dimension and innovation dimension, respectively, to obtain the traditional dimension score, aesthetic score and innovation dimension score, and the comprehensive evaluation result is obtained by weighting the scores of each dimension.

7. The method for generating traditional patterns and adapting them to industry based on AIGC as described in claim 1, characterized in that, When adapting the generated pattern images to industrial products, this includes layout adaptation, color adaptation, format conversion, and process parameter adaptation.

8. A traditional pattern generation and industry adaptation system based on AIGC, characterized in that, include: The image preprocessing module is used to acquire traditional pattern image data and preprocess the traditional pattern image data. The pattern generation module is used to generate traditional pattern images based on preprocessed traditional pattern image data and a trained pattern generation model; the construction process of the pattern generation model includes: A training dataset is constructed based on preprocessed traditional pattern image data. A noise prediction network is trained on the constructed training dataset to obtain a trained noise prediction network. The input control conditions are determined, and the trained noise prediction network is controlled according to the control conditions to generate traditional pattern images that meet the control conditions. A structured request is constructed, and a traditional pattern image optimization strategy is generated based on the structured request and a large language model. The traditional pattern images that meet the control conditions are optimized based on the generated traditional pattern image optimization strategy to obtain optimized traditional pattern images. The pattern modification module is used to determine the local area information to be modified in the optimized traditional pattern image, modify the local area information to be modified, and obtain the final traditional pattern image. The industry adaptation module is used to adapt the final traditional pattern image to industrial products to obtain an industrial product adaptation solution.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the traditional pattern generation and industry adaptation method based on AIGC as described in any one of claims 1-7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the traditional pattern generation and industry adaptation method based on AIGC as described in any one of claims 1-7.

Images