A collaborative constraint design and generation system and method for a multidimensional corpus of Yunjin brocade

Abstract

This invention discloses a design generation system and method based on collaborative constraints of a multi-dimensional corpus for Nanjing brocade. It constructs a multi-dimensional hierarchical corpus for Nanjing brocade, and labels multimodal data related to Nanjing brocade design at semantic, syntactic, and pragmatic levels, establishing cross-level association rules. Based on this, semantic feature vectors, syntactic constraint parameters, and pragmatic constraint parameters are extracted to construct core element feature vectors and a multi-level collaborative constraint parameter set. These are then introduced into a design generation engine for multi-level constraint control. A pragmatic feasibility verification mechanism performs closed-loop correction on the generated results until cultural semantic consistency, structural and technological rules, and application conditions are met before outputting the design scheme. This invention achieves collaborative control of semantic expression, structural rules, and application feasibility in the Nanjing brocade design generation process, ensuring the accuracy of cultural transmission and improving design efficiency and the feasibility of results.

Description

Technical Field

[0001] This invention relates to the fields of computer-aided design and digitalization of traditional processes, specifically to a design generation system and method for collaborative constraints of a multi-dimensional corpus of Nanjing brocade. Background Technology

[0002] As an important representative of traditional Chinese silk weaving, Nanjing brocade involves a variety of elements in its design process, including pattern structure, weaving techniques, color system, and cultural connotations, exhibiting significant systematicity and regularity. Traditional Nanjing brocade design mainly relies on the experience of artisans, with related cultural semantic information, process structural rules, and application scenario norms existing in a scattered manner, lacking a unified structured expression method, making it difficult to support efficient digital design generation.

[0003] With the development of digital and generative technologies, existing design assistance methods for Nanjing brocade mainly include creative generation methods based on general image generation models, computer-aided design software, and template-based rule-based combined generation systems. While these methods can generate patterns or design forms, their technical approaches primarily focus on visual feature expression or rule splicing, and therefore have the following problems:

[0004] (1) The cultural connotations, etiquette hierarchy and symbolic system carried by the brocade pattern were not structurally modeled, and the generated result lacked cultural semantic constraints; the brocade weaving structure type, process parameters and material constraints were not included in the generation control process, and the generated pattern may not conform to the actual weaving rules.

[0005] (2) The pattern was not systematically modeled for the conditions of use, production methods and suitable carriers in specific application scenarios, and there was no effective matching between the generated results and the actual application; there was no unified mechanism to coordinate and constrain cultural semantic information, structural process rules and application scenario conditions, and it was difficult to ensure the consistency of the generated results in multiple requirements.

[0006] In view of this, the present invention proposes a design and generation system and method for collaborative constraints of Yunjin multidimensional corpus. Summary of the Invention

[0007] The purpose of this invention is to provide a design generation system and method for multi-dimensional corpus collaborative constraints of Nanjing brocade, aiming to solve the technical problem in the prior art that it is impossible to perform unified structured modeling and collaborative constraints on the cultural semantic information, structural process rules and application scenario conditions of Nanjing brocade design, thus making it difficult to generate design schemes that simultaneously meet cultural accuracy, process feasibility and application adaptability.

[0008] In a first aspect, the present invention provides a method for designing and generating collaborative constraints for a multidimensional corpus of Yunjin brocade, characterized by comprising the following steps:

[0009] S1. Construct a multi-dimensional hierarchical corpus of Nanjing brocade, standardize the multimodal data related to Nanjing brocade design, and perform hierarchical annotation according to semantic, syntactic and pragmatic layers. Set a unique corpus identifier for each Nanjing brocade corpus object, so that the same corpus object forms cross-layer bound hierarchical annotation data at different levels.

[0010] S2. Based on the unique corpus identifier, establish cross-layer association rules between the semantic layer, syntactic layer and pragmatic layer to limit the legal combination paths between elements at different levels;

[0011] S3. Extract semantic feature vectors, syntactic constraint parameters and pragmatic constraint parameters from the hierarchical labeled data respectively, and map the three layers of features to a unified feature space to form core element feature vectors.

[0012] S4. Receive design requirements input by the user, perform hierarchical parsing of the design requirements, generate semantic feature vectors, syntactic constraint parameters and pragmatic constraint parameters, and perform legality screening of the constraint parameters based on the cross-layer association rules.

[0013] S5. Input the selected multi-level collaborative constraint parameter set into the design generation engine to control the direction of cultural expression at the semantic level, control the composition topology and color organization at the grammatical level, and control the process complexity and production condition range at the pragmatic level to generate preliminary design results.

[0014] S6. Perform feasibility verification on the preliminary design results based on pragmatic layer constraint parameters. If the preliminary design results do not meet the preset pragmatic layer constraint parameters, adjust the syntactic layer constraint parameters and re-execute the generation process. If the generation results simultaneously meet the consistency requirements of the semantic layer, syntactic layer, and pragmatic layer, output the design scheme.

[0015] As a preferred technical solution of the first aspect of the present invention, the cross-layer association rule uses a three-dimensional rule matrix to represent the combination relationship between semantic layer elements, syntactic layer elements and pragmatic layer elements. When the matrix value meets the preset conditions, it indicates that the combination is allowed; otherwise, it indicates that the combination is prohibited. The legality of the combination path of the three-layer elements is screened before generation.

[0016] As a preferred technical solution of the first aspect of the present invention, the logic for obtaining the semantic feature vector is as follows:

[0017] The semantic feature vector is obtained by semantically encoding the text data using a large language model, and is represented as E. sem =f LLM (T), where f LLM (*) indicates a semantic encoding function based on a large language model;

[0018] A semantic similarity algorithm is used to match the semantic feature vectors of design requirements with the semantic feature vectors of Yun brocade corpus objects. This is used to filter Yun brocade corpus objects that are consistent with the design requirements at the cultural and semantic level. The semantic similarity algorithm is as follows:

[0019] ; where: E sem i E represents the i-th semantic feature vector; sem j Sim represents the j-th semantic feature vector; sem (i,j) represents the semantic similarity between the i-th semantic feature vector and the j-th semantic feature vector; ||·|| represents the magnitude of the vector.

[0020] As a preferred technical solution of the first aspect of the present invention, structural topology recognition and parameterization calculation are performed on the pattern image to obtain contour node number parameters, line density parameters and color layer distribution parameters; after obtaining the semantic layer constraint set, matching grammatical rule entries are selected in the corpus according to the semantic layer constraint set; the selected grammatical rule entries are converted into grammatical layer constraint parameters to limit the composition topology, color layer organization method and weaving structure type in the design generation process.

[0021] As a preferred technical solution of the first aspect of the present invention, the pragmatic layer constraint parameters include application scenario category, adaptation carrier type, production method parameters and cost range parameters, and establish a threshold mapping relationship between the grammatical layer structural complexity index and the production capacity range, which is used to verify the feasibility of the grammatical layer constraint parameters.

[0022] As a preferred embodiment of the first aspect of the present invention, the logic for obtaining the core element feature vector includes:

[0023] The semantic feature vector, syntactic constraint parameters and pragmatic constraint parameters are used to screen the legality of the three-layer feature combination through cross-layer association rules, and feature combinations that do not meet the cultural etiquette rules or structural and technological constraints are eliminated.

[0024] The selected three-layer features are mapped to a unified feature space and fused according to preset weights to generate a core element feature vector. The core element feature vector is used to characterize the comprehensive expression state of the Yunjin corpus object in terms of cultural semantics, structural form and application attributes.

[0025] As a preferred embodiment of the first aspect of the present invention, the construction of the multi-level collaborative constraint parameter set includes:

[0026] The user-input design requirements are parsed into semantic feature vectors, syntactic constraint parameters, and pragmatic constraint parameters; the legality of the three constraint parameters is screened based on the cross-layer association rule matrix to form candidate constraint combinations;

[0027] The feasibility of the syntactic layer constraint parameters is verified based on the pragmatic layer production condition range. When the syntactic layer constraint parameters are detected to exceed the production capacity range of the target application scenario, the syntactic layer parameter adjustment mechanism is triggered to shrink the structural complexity index. The constraint combination that has been screened and verified is determined as a multi-level collaborative constraint parameter set to control the design generation path.

[0028] As a preferred embodiment of the first aspect of the present invention, the design generation engine adopts a conditional generation model. This conditional generation model takes the fused semantic feature vector, syntactic constraint parameters, and pragmatic constraint parameters as conditional inputs, and guides the generation process through a conditional control mechanism. Its core generation process is expressed as follows:

[0029] G=f SD (E sem E syn E pra );

[0030] Among them: E sem E represents a semantic feature vector. syn E represents the constraint parameters of the grammatical structure level. pra f represents pragmatic layer constraint parameters; SD (E sem E syn E pra () indicates the conditional control mechanism used by the design generation engine; G indicates the guiding generation result;

[0031] A low-rank adaptive parameter is introduced for fine-tuning, and the fine-tuning process satisfies:

[0032] W' = W + ΔW, ΔW = AB

[0033] Where W represents the original model parameters, and A∈R d×r B belongs to R r×k ΔW is a low-rank matrix, r≪min(d,k); ΔW is the low-rank weight increment matrix; W' is the fine-tuned model parameter, used to enable the model to learn the structural form and color organization rules of brocade patterns.

[0034] As a preferred embodiment of the first aspect of the present invention, the joint optimization objective function introduced in the design generation process is expressed as:

[0035] L=λ1L sem +λ2L syn +λ3L pra ;

[0036] Among them, L sem L represents the loss of cultural semantic bias. syn L represents the loss due to structural process deviation. pra The loss represents the mismatch between the application scenario and the target function, where λ1, λ2, and λ3 are weighting coefficients. When the joint optimization objective function exceeds a preset threshold, the design generation engine is triggered to re-execute the generation process.

[0037] Secondly, the present invention provides a design generation system based on collaborative constraints of a multidimensional corpus of Yunjin brocade, for running the first aspect, including:

[0038] The multidimensional hierarchical corpus construction module is used to collect and organize multimodal data related to Nanjing brocade design, and divide it into semantic, syntactic and pragmatic layers, and set a unique corpus identifier for each Nanjing brocade corpus object;

[0039] The hierarchical annotation module is used to perform semantic annotation, syntactic annotation and pragmatic annotation on the Yunjin corpus objects respectively, and to establish cross-layer association rules through the unique corpus identifier;

[0040] The core element feature extraction module is used to extract semantic feature vectors, syntactic constraint parameters and pragmatic constraint parameters respectively, and map the features to a unified core element feature vector.

[0041] The requirements parsing module is used to parse the user-input design requirements into semantic feature vectors, syntactic constraint parameters, and pragmatic constraint parameters.

[0042] The design generation engine is used to introduce the multi-level collaborative constraint parameter set into the design generation engine, perform multi-level constraint control on the generation process, and generate design results.

[0043] The pragmatic verification module is used to verify the pragmatic feasibility of the generated results and to adjust the generated parameters when the preset pragmatic layer constraint parameters are not met.

[0044] The multimodal design generation output module is used to output design schemes that meet the consistency requirements of the semantic, syntactic, and pragmatic layers.

[0045] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0046] This invention constructs a multi-dimensional hierarchical corpus based on semantic, syntactic, and pragmatic layers, and introduces a multi-layered collaborative constraint and consistency verification mechanism during the design and generation process. This enables cultural connotations, weaving structure rules, and application scenario conditions to be uniformly modeled and jointly controlled within the same technical framework. Since the design generation no longer relies solely on visual feature matching or template splicing, but is condition-driven and dynamically verified based on hierarchical structured data, the generated results can maintain accuracy in cultural expression, meet the requirements of weaving process parameters at the structural level, and conform to specific usage scenarios and production condition constraints at the application level. This improves the feasibility and consistency of the design scheme and reduces the time cost and experience dependence caused by repeated manual adjustments. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0048] Figure 1 This is a schematic diagram of the overall structure of the intelligent design generation system based on the Yunjin multidimensional hierarchical corpus of the present invention;

[0049] Figure 2 This is a schematic diagram of the annotation structure of the Yunjin semantic-structural-pragmatic multidimensional hierarchical corpus of the present invention;

[0050] Figure 3 This is a schematic diagram of the core element feature extraction and fusion process based on multidimensional hierarchical corpus of the present invention;

[0051] Figure 4 A schematic diagram of the semantic-syntactic-pragmatic multi-layer parsing modeling process for user design requirements of this invention;

[0052] Figure 5 This is a schematic diagram of the intelligent design generation and closed-loop verification process based on multi-level collaborative constraints of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings.

[0054] Throughout the accompanying drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions. The described embodiments are only a part of the embodiments of this application, not all of them. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0055] Example 1

[0056] like Figure 1 As shown, this embodiment provides a design generation system for collaborative constraints of Yunjin multidimensional corpus, including: a data acquisition module, a multidimensional hierarchical corpus construction module, a core element feature extraction module, a design generation engine module, and a multimodal design generation output module. The modules are connected sequentially according to the data flow to form a closed-loop processing flow from data acquisition to design output.

[0057] The data acquisition module is used to acquire multimodal data related to the design of Nanjing brocade, including image data, text data, structural parameter data, and application scenario data.

[0058] Specifically, the image data is acquired through high-definition scanning or photography equipment; the text data is obtained through document extraction or corpus crawling; the structural parameter data comes from weaving specifications and process data; and the application scenario data comes from historical cases and modern product application data. The acquired data is transmitted to a multi-dimensional hierarchical corpus construction module for hierarchical processing.

[0059] The multidimensional hierarchical corpus construction module divides the multimodal data into semantic, syntactic, and pragmatic layers. The semantic layer is used to express cultural connotations, hierarchical etiquette, and symbolic systems; the syntactic layer is used to express structural topology, process types, and material rules; and the pragmatic layer is used to express application scenarios, production conditions, and cost ranges.

[0060] Each Yun Brocade corpus object is assigned a unique corpus identifier, enabling the same Yun Brocade corpus object to form interconnected hierarchical labeled data at different levels. The Yun Brocade corpus object includes at least image data index, text data index, structural parameter index, and application scenario index.

[0061] Cross-layer association rules are established, storing image data, structured data, and feature vectors in corresponding data units. These rules, represented as a three-dimensional rule matrix, depict the combination relationships between semantic, syntactic, and pragmatic elements, and are used to filter the legality of the three-layer element combination paths in the subsequent generation stage. Specific judgment conditions are as follows:

[0062] Conditions for allowing combination: When the features to be combined meet the requirements of multi-level consistency, that is, when the cultural symbolism and etiquette level of the semantic layer, the structural topology and weaving technology of the grammatical layer, and the application scenario and production conditions of the pragmatic layer are mutually matched (for example, the semantics of "wealth and auspiciousness" corresponds to the central symmetric topology of "patterned peony" and the grammatical rule of "red and gold main color", and is legally applied to suitable pragmatic scenarios such as "wedding attire"), and when the structural complexity indicators do not exceed the production capacity and cost constraints of the target application scenario, the matrix values ​​meet the preset conditions, are judged to be allowed to combine, and a legal multi-level collaborative constraint parameter set is formed;

[0063] Conditions for prohibited combinations: When at least one conflict exists between elements at multiple levels, the combination is deemed prohibited and eliminated. Specific conflict scenarios include:

[0064] 1. Cultural regulatory conflict: The ritual hierarchy or sacred connotations carried by the semantic layer do not match the daily application scenarios or carriers of the pragmatic layer (such as specific patterns of sacrifice or imperial power being mistakenly used on ordinary daily necessities).

[0065] 2. Structural process exceeding limits: The structural complexity parameters such as the number of contour nodes, line density, and color layer distribution extracted from the grammatical layer exceed the weaving process limitations (such as the inability of machine weaving to achieve specific hand-colored shading techniques) or production capacity range defined by the pragmatic layer.

[0066] 3. Form and semantics do not match: The specific cultural images at the semantic level lack the legitimate structural topology or color organization rules at the grammatical level in the cross-layer association of the corpus (such as the color concept of the five primary colors being applied to modern color models that do not conform to regulations).

[0067] The core feature extraction module is used to extract computable core feature vectors from hierarchically labeled data; the core feature vectors include:

[0068] Semantic feature vectors are extracted by a semantic feature vector extraction unit to semantically encode text data. This unit employs a Large Language Model (LLM) to semantically understand and encode the textual descriptions and design requirements within the Yun Brocade corpus. The LLM is based on pre-trained language representation capabilities and adapted to the Yun Brocade domain corpus, enabling it to accurately understand the cultural connotations, symbolic meanings, and ritual attributes inherent in Yun Brocade patterns, and to map the text into semantic feature vectors.

[0069] For the input text T, the semantic feature vector is represented as:

[0070] E sem = f LLM (T), where f LLM (*) indicates a semantic encoding function based on a large language model.

[0071] In one implementation, a semantic similarity algorithm is used to match the semantic feature vectors of design requirements with the semantic feature vectors of Yun brocade corpus objects, thereby filtering Yun brocade corpus objects that are consistent with the design requirements at the cultural and semantic level. Semantic similarity is defined as:

[0072] ;

[0073] Among them: E sem i E represents the i-th semantic feature vector; sem j Sim represents the j-th semantic feature vector; sem (i,j) represents the semantic similarity between the i-th semantic feature vector and the j-th semantic feature vector; ||·|| represents the magnitude of the vector, thus forming a set of semantic feature vectors corresponding to the current design requirements, which is used to limit the cultural theme and symbolic meaning of the generated result.

[0074] After determining the set of semantic constraints, the syntactic layer performs structural mapping and rule matching on the design requirements. The syntactic layer describes the formal expression of cultural semantics in brocade patterns, including pattern type, structural organization, composition rules, color structure type, and weaving technique category. Based on the set of semantic constraints, corresponding syntactic rule entries are selected from the corpus and converted into syntactic layer constraint parameters to limit the range of possible design schemes at the structural level.

[0075] The grammatical constraint parameters, including the number of contour nodes, line density, and color layer parameters, are used to perform topological analysis calculations on the pattern structure based on the grammatical feature extraction unit. According to the semantic constraint set, corresponding grammatical rule entries are selected from the corpus and converted into grammatical constraint parameters to limit the range of possible design schemes at the structural level.

[0076] The pragmatic layer constraint parameters are generated by transforming application scenarios and production conditions into pragmatic layer labeled data based on the pragmatic feature extraction unit. Based on the pragmatic layer labeled data, the pragmatic feasibility of the current semantic feature vector and the syntactic layer constraint parameters is verified. When it is detected that the syntactic layer constraint parameters do not meet the production conditions or usage specifications of the target application scenario, the syntactic layer constraint parameters are adjusted or re-matched to form a multi-level collaborative constraint parameter set that meets the application requirements. When the generated result meets the consistency requirements of the semantic layer, syntactic layer, and pragmatic layer, the design scheme is output.

[0077] The extracted semantic feature vectors, syntactic constraint parameters, and pragmatic constraint parameters are mapped to a unified feature space to form core element feature vectors.

[0078] The design generation engine module is used to generate Yun Brocade design schemes based on hierarchical collaborative constraints. The design generation engine receives semantic feature vectors, syntactic constraint parameters, and pragmatic constraint parameters. During the generation process, the semantic feature vectors control the direction of theme expression; the syntactic constraint parameters limit the composition topology and color distribution; and the pragmatic constraint parameters limit the structural complexity and process load range. The generated result outputs preliminary design drawings.

[0079] The design generation engine employs a conditional generation model based on a diffusion mechanism, and the parameters of this generation model are fine-tuned for the Nanjing brocade design task. Based on pattern images, structural annotations, and semantic conditions in the Nanjing brocade multimodal corpus, low-rank adaptive parameters (LoRA) are introduced into the diffusion generation model. While keeping the basic model parameters frozen, only the low-rank parameters related to the Nanjing brocade pattern features are trained, so that the generation model can learn the unique pattern structure, color organization rules, and craft style characteristics of Nanjing brocade.

[0080] The LoRA fine-tuning process can be represented as:

[0081] W' = W + ΔW, ΔW = AB

[0082] Where W represents the original model parameters, and A∈R d×r B belongs to R r×k is a low-rank matrix, r≪min(d,k); ΔW is the low-rank weight increment matrix; W' is the fine-tuned model parameters.

[0083] During the generation process, the semantic tags corresponding to the semantic layer in the multi-level collaborative constraint parameter set are converted into semantic control conditions and introduced into the generation model as conditional inputs to limit the cultural themes, symbolic objects, and ritual attributes expressed in the generated results. The structural parameters corresponding to the grammatical layer in the multi-level collaborative constraint parameter set are used as grammatical layer control conditions to constrain the expression of pattern composition, structural organization, and weaving process characteristics during the generation process, ensuring the legitimacy of the generated results at the formal structure and technological implementation levels.

[0084] The multimodal design generation output module is used to organize and encapsulate the generated results. The output includes: brocade pattern image files, color encoding parameters, structural and process parameter descriptions, semantic source path records, and application adaptation specification files. When the generated results do not meet pragmatic constraints, the design generation engine will be triggered to re-execute the generation process until the preset conditions are met.

[0085] During the generation process, pragmatic layer constraints are continuously applied to the generated results. These constraints limit the usability and manufacturability of the generated results in the target application scenario. When the generated results do not meet the pragmatic layer constraints, the generation parameters are dynamically adjusted, including reducing structural complexity, adjusting process parameters, or recombining syntactic layer rules, until the generated results meet the application requirements. During or after generation, a multi-objective constraint optimization algorithm is introduced to jointly evaluate the matching degree of the generated results at the semantic, syntactic, and pragmatic layers, and optimizes the selection of the generated results based on a weighted loss function.

[0086] The joint optimization objective function is expressed as:

[0087] L=λ1L sem +λ2L syn +λ3L pra ;

[0088] Among them, L sem L represents the loss of cultural semantic bias. syn L represents the loss due to structural process deviation. pra λ1, λ2, and λ3 represent the application scenario mismatch loss, and λ3 are weight coefficients.

[0089] When the joint optimization objective function L exceeds the threshold, the design generation engine is triggered to regenerate.

[0090] During the generation process, the diffusion model uses the fused semantic feature vector, syntactic constraint parameters, and pragmatic constraint parameters as conditional inputs, and guides the generation process through a conditional control mechanism. Its core generation process is represented as follows:

[0091] G=f SD (E sem Esyn E pra );

[0092] Among them: E sem E represents a semantic feature vector. syn E represents the constraint parameters of the grammatical structure level. pra f represents pragmatic layer constraint parameters; SD (E sem E syn E pra ) indicates the condition control mechanism used by the design generation engine; G indicates the guidance generation result.

[0093] After the pattern is generated, the pragmatic feasibility of the generated result is verified. The verification includes, but is not limited to, whether the process complexity exceeds the set threshold, whether the number of colors meets the weaving process restrictions, and whether it conforms to the etiquette norms and usage requirements of the corresponding application scenario.

[0094] Once the verification is successful, a complete design scheme will be output. The design scheme will include at least the brocade pattern, color scheme description, corresponding weaving process parameters, and material list, which will be used to guide actual production or subsequent design applications.

[0095] This embodiment constructs a multimodal, multi-layered corpus for Nanjing brocade design, realizing the systematic expression of design source data. By introducing semantic, syntactic, and pragmatic layers, and combining multimodal data expression forms such as text, images, and structured parameters, it uniformly annotates and manages the cultural semantics, structural rules, and application scenarios related to Nanjing brocade design. This provides clear semantic basis, structural constraints, and pragmatic verification mechanisms for the design generation process, thereby improving design efficiency, enhancing the interpretability of the design process, and ensuring the cultural accuracy of the design results.

[0096] By simultaneously introducing semantic feature vectors, syntactic constraints, and pragmatic constraints into the design generation process, the generated scheme can be synergistically controlled in terms of cultural expression, structural rationality, and application feasibility, thus avoiding design deviations caused by relying solely on a single modality or a single level of control.

[0097] By using a multimodal corpus hierarchical annotation and feature mapping mechanism, the generated design schemes can clearly correspond to their semantic layer cultural semantics, grammatical layer structural rules, and pragmatic layer application scenario conditions. This improves the traceability, standardization, and controllability of the design results, making every design detail verifiable and facilitating the accumulation and reuse of design experience.

[0098] The design generation engine parses user-input design requirements into multi-layered design constraint parameters at the semantic, syntactic, and pragmatic levels. Based on this, it intelligently generates design schemes that meet the requirements of Nanjing brocade craftsmanship and application scenario limitations, achieving intelligent integration of the entire design process, significantly improving design efficiency, and reducing over-reliance on the experience of individual craftsmen.

[0099] Example 2

[0100] like Figure 2-5 As shown, this embodiment, based on embodiment 1, provides a method for designing and generating collaborative constraints for the Yunjin multidimensional corpus, including the following steps:

[0101] A multi-dimensional hierarchical corpus of Nanjing brocade was constructed. Multimodal data related to Nanjing brocade design were collected and organized. The multimodal data was divided into semantic, syntactic and pragmatic layers. A unique corpus identifier was set for each Nanjing brocade corpus object, so that the same Nanjing brocade corpus object formed interrelated hierarchical labeled data at different levels.

[0102] Specifically, based on the "semantic-syntactic-pragmatic" architecture of Nanjing Brocade, multiple data collection methods are used to collaboratively collect and organize multimodal data such as text data, image data, structured parameter data, and application case data related to Nanjing Brocade design, in order to construct a standardized multidimensional hierarchical corpus of Nanjing Brocade that supports multi-layered expression.

[0103] At the semantic layer, textual data was collected from publicly available documents such as historical materials from the Nanjing Yunjin Museum, official brand statements, and the "Complete Collection of Chinese Traditional Crafts" through text crawling and text extraction technologies. Combined with interview records of Yunjin inheritors, semantic elements such as symbolic meanings, cultural connotations, and etiquette levels corresponding to Yunjin patterns were extracted and organized to establish calculable semantic layer data, ensuring the authority of the data sources and the cultural accuracy of the design basis.

[0104] At the linguistic layer, high-resolution image data is obtained by scanning and photographing the exquisite Nanjing brocade artifacts and related weaving tools. Annotation tools are then used to annotate the pattern structure features, composition methods, and technological elements. At the same time, weaving parameter information recorded in patent documents and product specifications is integrated to transform weaving process requirements such as "brocade patterns" into quantifiable structural parameters and process constraints, thereby forming linguistic layer data. This transforms the originally implicit process experience into explicit structured data, providing support for the technological feasibility of the design scheme.

[0105] At the pragmatic level, we organized application cases of Nanjing brocade patterns in different historical periods and on different media, including ancient ceremonial clothing, modern clothing and cultural and creative products, and extracted pragmatic information related to usage scenarios, adapted media and production conditions to construct pragmatic layer data of Nanjing brocade patterns in different application contexts.

[0106] Through the collection and fusion of the aforementioned multi-source heterogeneous and multimodal data, based on a three-layer structure of "semantic layer—structural layer—pragmatic layer," semantic information, structural and technological information, and application scenario information related to Nanjing brocade design are organized within a unified framework. The corpus related to Nanjing brocade design is collected and organized in a layered manner, and its layered system includes the following:

[0107] The semantic layer is used to express the cultural symbolism and value system carried by the patterns of Nanjing brocade. Its corpus includes the following categories:

[0108] Symbolic and allegorical corpora include those representing auspicious wishes and those symbolizing protection from harm; among them:

[0109] Auspicious wishes include themes such as praying for children and longevity, receiving blessings and wealth, and good fortune. Praying for children and longevity corresponds to patterns such as phoenixes among peonies, fish playing among lotuses, pomegranates bursting with seeds, numerous offspring, pine trees and cranes symbolizing longevity, and tortoises and cranes symbolizing long life. Receiving blessings and wealth corresponds to patterns such as fish leaping over the dragon gate, congratulations on getting rich, and abundant harvests. Good fortune corresponds to patterns such as joy on eyebrows, bamboo bringing peace, the arrival of spring, and continuous abundance year after year.

[0110] Symbolic patterns for warding off harm include semantic elements related to exorcising evil and avoiding disasters, involving mythical beast patterns (such as the Four Gods, Qilin, and tortoise), images of immortals and warriors (such as Guanyin, the Eight Immortals, and Zhong Kui), and symbolic patterns related to the Five Poisons (such as scorpion, toad, snake, centipede, and spider).

[0111] Folk culture corpus includes folk customs and concepts, which in turn includes corpus related to folk customs and sentiments.

[0112] The religious and philosophical texts include concepts of the universe, the relationship between heaven and humanity, and the divine right of kings. Among them, the concept of the universe includes the idea of ​​"continuous creation"; the concept of the relationship between heaven and humanity includes "the unity of heaven and humanity"; and the concept of the divine right of kings includes the idea of ​​the divine right of kings.

[0113] The corpus on ethnic beliefs includes forms of worship, which include cultural content such as totem worship and red and yellow worship.

[0114] The corpus of etiquette hierarchy includes institutional functions, social distinctions, control and identification. Among them, institutional functions include etiquette system tools, social distinctions are made by different colors or patterns to distinguish social ranks, and control and identification are achieved by limiting colors to realize identity recognition and control functions.

[0115] The grammatical layer is used to express the formal rules of Nanjing brocade patterns in terms of materials, structure, craftsmanship, and color organization. Its corpus includes:

[0116] Material symbols include silk, metal, and decorative natural materials. Silk includes silkworm silk; metal includes gold and silver threads, specifically round gold thread, sheet gold thread, and gold leaf thread; and decorative natural materials such as peacock feathers are also included.

[0117] The categories of craft symbols and weaving structures include gold-woven fabrics (such as gold and silver brocades), brocades (such as colorful brocades, Tianhua brocades, gold-woven brocades, two-color gold brocades, hibiscus brocades, gold-colored velvet, etc.), satin (such as gold-embroidered satin, natural-colored brocades, two-color brocades with ground patterns, brocaded satin, and imperial robes), and decorative fabrics (such as gold-embroidered ground fabrics, decorative satin, decorative yarn, decorative silk, and decorative brocade).

[0118] The weaving process includes design preparation, pattern selection and binding (such as pattern selection, reverse pattern, and patchwork), machine making process (machine setup, pattern making and weft fabrication, warp loading, yarn counting, fiber drawing, yarn picking, reed threading, etc.), and weaving process (pattern pulling, coil weaving, warp cutting and weft breaking, weft cutting through tubes, etc.).

[0119] Weaving tools include large jacquard looms, small jacquard looms, and auxiliary tools such as shuttles, weft knives, and weft tubes.

[0120] Color symbols include color concepts, color classifications, color systems, and color application techniques. Specifically: color concepts include the five primary colors (cyan, red, yellow, white, and black) and the concept of red and yellow as the primary colors; color classifications include warm colors (red, bright yellow, gold, pale yellow, pink, beige, orange, eggplant purple, and light purple, etc.), neutral colors (white and silver), and cool colors (green, lake blue, lapis lazuli, magenta, and moon white, etc.); color systems include the red-orange system (such as scarlet, true red, vermilion, silver red, and light red, etc.), the yellow-green system (such as true yellow, bright yellow, acacia yellow, and golden yellow, etc.), and the blue-purple system (such as sea blue, royal blue, turquoise, and navy blue, etc.); and color application techniques include color halos (such as two-color halos, three-color halos, positive halos, reverse halos, and alternating colors), techniques like alternating white and gold trim, etc.

[0121] The category of decorative symbols includes decorative themes, compositional formats, and rules for drawing decorative patterns. Decorative themes include plant motifs (such as lotus, peony, and lotus flower), animal motifs (such as dragon, phoenix, qilin, crane, bat, and fish), mythical treasures, human figures, textual patterns, and abstract symbols and natural objects. Compositional formats include clusters, scattered flowers, full blooms, intertwined branches, continuous branches, broken branches, and brocade patterns. Rules for drawing decorative patterns include general principles for flowers, and compositional rules for drawing dragons, phoenixes, intertwined lotus, peonies, bats, and clouds.

[0122] The pragmatic layer is used to express the functional attributes and application conditions of brocade patterns in actual use contexts, and its corpus content includes:

[0123] Use functional categories, including clothing applications (official robes, formal wear, everyday clothes, etc.), accessory applications (clothing accessories, etc.), home furnishing applications (ornaments, paintings, etc.), and cultural and creative derivative applications (refrigerator magnets, coasters, etc.).

[0124] Use case categories include scenarios related to ancient etiquette (wearing court robes, sacrificial activities, religious ceremonies, etc.), celebratory ceremonies (birthdays, graduations, business openings, housewarmings, weddings, etc.), and daily life scenarios (home decoration, etc.).

[0125] Through the hierarchical collection and organization of semantic, syntactic, and pragmatic layers, the cultural significance, structural and technological rules, and application scenarios related to Nanjing brocade design are integrated into a structured data system within a unified framework, providing a data foundation for subsequent hierarchical annotation, feature extraction, and collaborative generation.

[0126] Semantic annotation, syntactic annotation, and pragmatic annotation are performed on the Yunjin corpus objects respectively. Semantic annotation is used to describe cultural connotations and symbolic information, syntactic annotation is used to describe pattern structure, color rules and process parameters, and pragmatic annotation is used to describe application scenarios and usage conditions. Cross-layer association rules are established through the unique corpus identifier.

[0127] Specifically, the collected raw multimodal data needs to undergo standardization and hierarchical annotation to transform it into structured information units that are understandable, computable, and callable by computers. This step is crucial for building a multidimensional hierarchical corpus of Yunjin brocade with the capabilities of "cultural semantic traceability, reusable structural and technological parameters, and matching application scenario conditions."

[0128] The text data is deduplicated, standardized in terminology, and segmented, and the text content is converted into a vectorized representation using a semantic coding model; the image data is standardized in format, noise is removed, and key regions are labeled; the structured parameter data is normalized and validated to ensure the consistency and reliability of subsequent multi-layer labeling and computation.

[0129] To achieve unified management and multi-layered structured representation of Nanjing brocade corpus objects, the Nanjing brocade corpus objects and their corresponding hierarchical annotation units are clearly defined and coded. A single Nanjing brocade corpus object refers to a complete instance of a Nanjing brocade corpus object that can be used for design analysis and generation. It includes at least one image of a Nanjing brocade fabric and its corresponding text description information. The text description information includes the name of the Nanjing brocade pattern, its cultural connotation, its craftsmanship type, and its historical usage scenarios. Each Nanjing brocade corpus object serves as the basic carrier for multi-dimensional structured annotation and corresponds to only one unique corpus identifier at the corpus level.

[0130] When performing multimodal hierarchical annotation on the Yunjin (Yunnan brocade) corpus, this invention divides the annotated content into three categories: semantic layer, syntactic layer, and pragmatic layer. The semantic layer defines the cultural semantic expression boundaries of the design content; the syntactic layer defines the formal structure and technological implementation boundaries of the design content; and the pragmatic layer defines the usability and production feasibility boundaries of the design content in specific application scenarios. Simultaneously, independent annotation units are generated for each layer, thereby ensuring that the corpus can both express the intangible cultural heritage connotations and support technological feasibility analysis and application scenario matching.

[0131] At the semantic level, the focus is on depicting the cultural symbolism and value system carried by the patterns and decorative elements of Nanjing brocade. The semantic annotation content includes symbolic meaning, folk cultural connotations, religious philosophical concepts, national belief attributes, and information on ritual hierarchy. Through the above semantic annotation, the cultural connotations corresponding to the Nanjing brocade patterns can be transformed into calculable and searchable semantic units, which participate in the design selection and combination as cultural consistency constraints during the design generation process, ensuring the cultural accuracy of the design scheme from the source.

[0132] At the grammatical level, the focus is on describing the structural characteristics of Nanjing brocade patterns in terms of craftsmanship and form. The grammatical annotations include weaving process elements, material elements, color elements, and decorative elements. By structurally annotating these grammatical level elements, Nanjing brocade patterns are transformed from descriptions relying on visual experience into a set of structural rules that can be invoked by design systems and constrain the generation process. This facilitates the rapid reuse of process parameters and improves design conversion efficiency.

[0133] At the pragmatic level, the focus is on reflecting the functional attributes and application conditions of Nanjing brocade in social use. The pragmatic annotation content includes usage functions, usage scenarios, suitable carriers, user groups, and usage restrictions. Through pragmatic layer annotation, the multi-dimensional hierarchical corpus of Nanjing brocade can be directly matched with the application conditions in specific design requirements, thereby improving the feasibility and adaptability of the design results in actual use scenarios.

[0134] Based on the three types of annotation content mentioned above, this invention integrates semantic, syntactic, and pragmatic elements into a unified structured annotation system, and achieves cross-layer binding through preset cross-layer association rules, thereby forming a complete multi-dimensional structured record of Nanjing brocade corpus objects. This record not only preserves the integrity of the traditional cultural symbol system of Nanjing brocade, but also meets the needs of computers for analyzing, retrieving, and controlling the generation of design elements.

[0135] Based on the hierarchical labeled data, semantic feature vectors, syntactic constraint parameters and pragmatic constraint parameters are extracted respectively, and the features are mapped to a unified core element feature vector; the user input design requirements are received, and the design requirements are parsed into semantic feature vectors, syntactic constraint parameters and pragmatic constraint parameters;

[0136] Specifically, to distinguish between different levels of annotation units, a unified encoding rule is adopted for each level of annotation units. Semantic layer annotation units are prefixed with "SEM-", syntactic layer annotation units are prefixed with "SYN-", and pragmatic layer annotation units are prefixed with "PRA-". Subsequent characters are used to distinguish the specific annotation object and its sequence number.

[0137] In this embodiment, the Yun brocade corpus objects are identified using a unified numbering rule, where "CJN" represents a Yun brocade corpus object, "EX" represents an example corpus, and "001" represents the first Yun brocade design corpus selected in this embodiment. It should be noted that while each Yun brocade corpus object corresponds to only one unique corpus identifier at the corpus level, three annotation units—semantic, syntactic, and pragmatic—can be generated simultaneously at the annotation level. These annotation units are all associated and bound through the corpus identifier, thus forming a complete multi-layered structured Yun brocade corpus object record.

[0138] The following example uses the Yun Brocade corpus object numbered CJN-EX-001. This Yun Brocade corpus object has its semantic element identifier set as "SEM-001" in the semantic layer. The corresponding element name is "Peony Pattern with Floral Design." Its auspicious meaning is labeled as "wealth and good fortune," and its cultural origin is labeled as "Ming and Qing Dynasty court etiquette." Regarding etiquette constraints, this pattern is labeled as suitable for use in wedding and investiture scenarios. Associated semantic elements include "pomegranate (many seeds)" and "bat (good fortune)," used to form semantic combination expansions or semantic enhancement matching during the design generation process. Through the above semantic layer annotation, the pattern forms a structured expression in terms of cultural meaning, historical origin, and etiquette usage, providing a data foundation for subsequent semantic matching and conflict detection.

[0139] The Yun Brocade corpus object is labeled with the syntactic element identifier "SYN-001" in the syntactic layer. Its decorative motif is labeled "plant pattern." Its craft type is labeled "brocade pattern." Its structural feature is labeled "centrally symmetrical layout in a floral pattern," used to limit the compositional topology. Its color rule is labeled "primarily red and gold, highly saturated," used to limit the main color tone and color intensity range during the generation process. Its material constraint is labeled "silkworm silk + gold thread," used to limit the available material combinations. Through these syntactic layer annotations, the pattern's structural form, craft type, and material constraints are formed into calculable structured parameters, providing a basis for compositional control and craftsmanship feasibility assessment during the generation stage.

[0140] The Yun brocade corpus object is labeled with the pragmatic element identifier "PRA-001" at the pragmatic layer. Its application scenario is labeled "wedding attire." Its suitable carrier is labeled "formal wear fabric." Its manufacturing process is labeled "hand-woven, low-volume production." Its target audience is labeled "traditional formal wear users." Its cost range is labeled "high-cost." Through these pragmatic layer annotations, the pattern is given clear constraints in terms of actual application scenarios, production methods, and target users, providing a basis for pragmatic feasibility verification and adaptation judgment after design generation.

[0141] In this embodiment, the semantic layer element "SEM-001", the syntactic layer element "SYN-001", and the pragmatic layer element "PRA-001" are associated and stored through a unique corpus identifier. When the design and generation system calls this corpus object, it can simultaneously read its cultural connotation information, structural process parameters, and application scenario conditions, thereby realizing the collaborative constraints between semantic expression, structural implementation, and usage adaptation.

[0142] The semantic feature vector, syntactic constraint parameters, and pragmatic constraint parameters are matched with the multi-level collaborative constraint parameter set to form a multi-level collaborative constraint parameter set for design generation. During the design generation process, the multi-level collaborative constraint parameter set is introduced into the design generation engine to perform multi-level constraint control on the generation process and generate preliminary design results.

[0143] Specifically, the processed and hierarchically labeled multimodal data of Nanjing Brocade is managed using a multi-database collaborative storage architecture to support efficient and accurate retrieval and retrieval under semantic, syntactic, and pragmatic constraints. To support the unified management of Nanjing Brocade corpus objects within the "semantic-syntactic-pragmatic" multi-layered annotation system and the subsequent design scheme generation process, this invention organizes the standardized and hierarchically labeled data using a multi-type data collaborative storage method. This achieves consistent maintenance of multimodal data, multi-layered constraint matching, and rapid retrieval, significantly improving the systematic nature of data management and retrieval efficiency.

[0144] In one embodiment of the present invention, the Yun Brocade corpus object uses a semantic layer, a syntactic layer, and a pragmatic layer as the core organizational hierarchy to express and store multi-level and multi-modal data related to Yun Brocade patterns. The Yun Brocade corpus object data includes at least image data, text data, structured parameter data, and corresponding feature vector data, which are used to uniformly carry and compute cultural semantic information (semantic layer), structural and technological information (syntactic layer), and application context information from different data representation forms.

[0145] To address the characteristics of different data types, appropriate data storage methods are employed for management, including:

[0146] Image data storage unit: used to store images of brocade patterns, process diagrams and their corresponding multi-layer annotation files. The image data can be managed using object storage to support unified access and retrieval of large-scale image data, and to provide a data foundation for subsequent analysis, matching and design generation based on visual features.

[0147] Structured data storage unit: used to store semantic fields, syntactic parameters, pragmatic constraints and their association rules generated during multidimensional annotation. The structured data is organized in relational or equivalent data tables to support precise query, mapping and combination operations based on semantic conditions, structural rules and usage scenario constraints.

[0148] Vector data storage unit: used to store feature vector representations extracted from Yunjin text corpus and image corpus. The feature vectors are used to calculate and express semantic layer information, syntactic layer constraint parameters and pragmatic layer similarity to support fast retrieval, matching and recommendation based on similarity.

[0149] The above-mentioned different types of data are linked through a unified and unique corpus identifier, so that the multimodal representation of the same brocade pattern at the semantic, syntactic and pragmatic levels can form a unified data unit, thereby constituting a brocade corpus object structure that can be collaboratively called by the designed system.

[0150] Based on the aforementioned multimodal, multi-level data storage architecture, this invention further provides a retrieval mechanism that matches the semantic-syntactic-pragmatic annotation system, used to support the selection and combination of design elements during the design scheme generation process. The retrieval mechanism can match user-inputted design requirements information from at least one of the following levels:

[0151] Semantic layer matching: Matching is performed based on cultural connotations, symbolic meanings, and etiquette attributes to ensure that the selected patterns meet the requirements of cultural expression and avoid cultural misuse;

[0152] Structural layer matching: Based on process type, structural parameters, material constraints, etc., matching is performed to select structural materials that meet the target process requirements;

[0153] Pragmatic matching: Matching is performed based on application scenarios, adaptation carriers, production conditions, etc., to ensure that the materials meet the needs of actual application scenarios.

[0154] In one implementation, a conditional filtering operation is first performed based on structured annotation fields to narrow down the candidate corpus; then, by combining feature vector similarity calculation, the candidate brocade corpus objects are sorted or further filtered, thereby returning a set of brocade corpus objects that meet multiple constraints at the semantic, syntactic, and pragmatic levels.

[0155] Through the above-mentioned data storage and retrieval management methods, this invention can achieve: unified management and dynamic maintenance of multimodal and multi-layered annotated corpora of Nanjing brocade, solving the problems of traditional data being scattered and difficult to reuse; efficient and multi-dimensional matching between design requirements and Nanjing brocade corpus objects, replacing the tedious process of manually searching for materials; and provide a stable, traceable, and layered data foundation for the subsequent extraction of core elements of Nanjing brocade and the generation of design schemes, ensuring the cultural accuracy and technological feasibility of the design output.

[0156] The semantic feature vector, syntactic constraint parameters, and pragmatic constraint parameters are matched with the multi-level collaborative constraint parameter set to form a multi-level collaborative constraint parameter set for design generation. During the design generation process, the multi-level collaborative constraint parameter set is introduced into the design generation engine to perform multi-level constraint control on the generation process and generate preliminary design results.

[0157] Specifically, from the multimodal corpus of Nanjing brocade with completed semantic-structural-pragmatic layered annotation, core element features that simultaneously possess cultural semantic expression capabilities, technological structural constraints, and application scenario adaptability are extracted and converted into core element feature vectors that can be called by the design generation engine. At the semantic layer, for the layered annotation data related to cultural connotations, symbolic meanings, and ritual attributes in the Nanjing brocade corpus objects, semantic feature vectors used to represent the cultural connotations of Nanjing brocade are extracted. Semantic encoding processing is performed on the corresponding text data in the Nanjing brocade corpus objects to generate semantic representations that reflect basic semantic content. Furthermore, combined with pre-constructed cross-layered association rules of Nanjing brocade semantic-structural-pragmatic, the semantic representations are integrated with cultural tags, symbolic meaning categories, and ritual level information to form semantic feature vectors used to represent the cultural semantic attributes of Nanjing brocade. The extracted semantic feature vectors not only reflect the surface-level linguistic meanings of the patterns but also depict their symbolic meanings, value orientations, and usage norms in the Nanjing brocade cultural system. In the design generation process, they participate in feature selection and combination as cultural consistency constraints.

[0158] At the grammatical level, based on the formal structure, technological rules, and visual composition information of the brocade pattern, grammatical features used to characterize the weavability and formal features of the pattern are extracted. Visual feature encoding is performed on the brocade pattern image to obtain basic visual features reflecting the overall shape of the pattern. Furthermore, structural lines, compositional outlines, and color layer distribution information in the pattern are identified to describe the grammatical features such as the layout of the floral patterns, the characteristics of the branching direction, and the hierarchical structure of the color halo. Cluster analysis is performed on the color information of the pattern to extract parametric representations corresponding to color structures such as "two-color halo" and "three-color halo" in the brocade weaving process, so as to reflect the technological organization of the pattern at the color level.

[0159] Through the above-mentioned process of extracting structural constraint parameters, the brocade pattern is transformed from a visual description method that relies on human experience to a set of structural constraint parameters that can be called by the design system and can form structural constraints on the design generation process.

[0160] At the pragmatic layer, pragmatic features are extracted to characterize the actual use and production feasibility of brocade patterns, considering their application scenarios, carrier adaptation conditions, and production process limitations. The usage scenario information, adapted carrier types, production methods, and process limitation parameters labeled in the corpus are mapped into pragmatic feature representations, and a correlation is established with the structural parameters in the grammatical layer to describe the usability and adaptability of the pattern under specific application conditions. By introducing the process parameter threshold in the grammatical layer, the structural complexity of the pattern is verified. When a feature combination is detected to exceed the preset process feasibility range, the corresponding structural features are adjusted or simplified to ensure that the generated feature representation meets the actual weaving and application conditions.

[0161] After extracting the semantic, syntactic, and pragmatic features, the features at each layer are uniformly represented and fused. The semantic feature vectors, syntactic features, and pragmatic features are mapped to a unified feature space and fused according to a preset weight ratio to generate a core element feature vector of Yun brocade that contains cultural semantic expression, structural rule constraints, and application scenario adaptation information. By introducing a consistency verification mechanism, it is ensured that there are no conflicts between the fused features in terms of cultural semantics, craft structure, and usage scenario, thereby generating a core element feature vector that conforms to the traditional norms of Yun brocade, which serves as the input for the subsequent design generation engine.

[0162] In this embodiment, the design generation engine does not generate Nanjing brocade patterns without constraints based on a single generation model. Instead, it controls the generation process and verifies the results under the combined effect of hierarchical knowledge and constraints of semantics, syntacticity, and pragmatics. This achieves the generation of design schemes that both conform to the connotation of intangible cultural heritage and meet the feasibility and application requirements of Nanjing brocade craftsmanship. Before generating the design scheme, the user-input design requirements are first structurally analyzed and element-decomposed, transforming them into a calculable and constrainable multi-level collaborative constraint parameter set, providing a unified constraint parameter foundation for subsequent design scheme generation.

[0163] The multi-level collaborative constraint parameter set includes three levels: semantic layer, syntactic layer, and pragmatic layer, which are used to analyze and express design requirements from three aspects: cultural semantics, structural rules, and application scenarios, respectively. The semantic layer is used to identify the cultural semantic information contained in the design requirements. It receives natural language design requirements input by the user, parses them using a semantic understanding model, and extracts cultural semantic elements, including but not limited to auspicious meanings, symbolic objects, ritual attributes, and cultural style types. The semantic feature vector extraction unit uses a Large Language Model (LLM) to perform semantic understanding and encoding on the textual description information and design requirement text in the Yun Brocade corpus. The LLM is based on pre-trained language representation capabilities and adapted to the Yun Brocade domain corpus, enabling it to accurately understand the cultural connotations, symbolic meanings, and ritual attribute information contained in Yun Brocade patterns, and to map the text into semantic feature vectors.

[0164] For the input text T, the semantic feature vector is represented as:

[0165] E sem = f LLM (T), where f LLM (*) indicates a semantic encoding function based on a large language model.

[0166] In one implementation, a semantic similarity algorithm is used to match the semantic feature vectors of design requirements with the semantic feature vectors of Yun brocade corpus objects, thereby filtering Yun brocade corpus objects that are consistent with the design requirements at the cultural and semantic level. Semantic similarity is defined as:

[0167]

[0168] Among them: E sem i E represents the i-th semantic feature vector; sem j Sim represents the j-th semantic feature vector; sem (i,j) represents the semantic similarity between the i-th and j-th semantic feature vectors; ||·|| represents the magnitude of the vector.

[0169] This forms a set of semantic feature vectors corresponding to the current design requirements, which are used to define the cultural theme and symbolic meaning of the generated results.

[0170] After determining the set of semantic constraints, structural mapping and rule matching of design requirements are performed at the grammatical layer. The grammatical layer describes the formal expression of cultural semantics in brocade patterns, including pattern type, structural organization method, composition rules, color structure type, and weaving technique category. Based on the set of semantic constraints, corresponding grammatical rule entries are selected from the corpus and converted into grammatical layer constraint parameters to limit the range of possible design schemes at the structural level.

[0171] The pragmatic feasibility of the preliminary design results is verified based on pragmatic layer constraint parameters. When the preliminary design results are found to not meet the preset pragmatic layer constraint parameters, the generation parameters are adjusted and the design generation process is re-executed. When the generation results meet the consistency requirements of semantic layer, syntactic layer and pragmatic layer, the design scheme is output.

[0172] Specifically, the pragmatic layer is used to verify the feasibility of design requirements in specific application scenarios. The pragmatic layer includes information such as application scenario type, adapted carrier form, production process conditions, and usage specifications.

[0173] Based on pragmatic layer annotation data, pragmatic feasibility is verified between the current semantic feature vector and the syntactic layer constraint parameters. When it is detected that the syntactic layer constraint parameters do not meet the production conditions or usage specifications of the target application scenario, the syntactic layer constraint parameters are adjusted or re-matched to form a multi-level collaborative constraint parameter set that meets the application requirements.

[0174] Through the above-mentioned hierarchical parsing and verification process, a complete and structured set of semantic-syntactic-pragmatic multi-level collaborative constraint parameters is obtained, which serves as the input conditions for the subsequent design and generation engine.

[0175] In this embodiment, the design generation engine executes the design scheme generation process based on the obtained multi-level collaborative constraint parameter set. This generation process does not rely on a single generation model for free generation, but is completed under the joint control of multiple constraint conditions.

[0176] During the generation process, the semantic tags corresponding to the semantic layer in the multi-level collaborative constraint parameter set are converted into semantic control conditions and introduced into the generation model as conditional inputs to limit the cultural themes, symbolic objects, and ritual attributes expressed in the generated results. The structural parameters corresponding to the grammatical layer in the multi-level collaborative constraint parameter set are used as grammatical layer control conditions to constrain the expression of pattern composition, structural organization, and weaving process characteristics during the generation process, ensuring the legitimacy of the generated results at the formal structure and technological implementation levels.

[0177] During the generation process, the diffusion model uses the fused semantic feature vector, syntactic constraint parameters, and pragmatic constraint parameters as conditional inputs, and guides the generation process through a conditional control mechanism. Its core generation process is represented as follows:

[0178] G=f SD (E sem E syn E pra );

[0179] Among them: E sem E represents a semantic feature vector. syn E represents the constraint parameters of the grammatical structure level. pra f represents pragmatic layer constraint parameters; SD (E sem E syn E pra ) indicates the condition control mechanism used by the design generation engine; G indicates the guidance generation result.

[0180] A conditional generative model based on a diffusion mechanism is adopted, and the parameters of the generative model are fine-tuned for the Nanjing brocade design task. Based on the pattern images, structural annotations, and semantic conditions in the Nanjing brocade multimodal corpus, low-rank adaptive parameters (LoRA) are introduced into the diffusion generative model. While keeping the basic model parameters frozen, only the low-rank parameters related to the Nanjing brocade pattern features are trained, so that the generative model can learn the unique pattern structure, color organization rules, and craft style features of Nanjing brocade.

[0181] The LoRA fine-tuning process can be represented as:

[0182] W' = W + ΔW, ΔW = AB;

[0183] Where W represents the original model parameters, and A∈R d×r B belongs to R r×k is a low-rank matrix, r≪min(d,k); ΔW is the low-rank weight increment matrix; W' is the fine-tuned model parameters.

[0184] After the generated pattern is completed, a pragmatic feasibility check is performed on the generated result. The pragmatic constraints are used to limit the usability and manufacturability of the generated result in the target application scenario. When the generated result does not meet the pragmatic constraints, the generation parameters are dynamically adjusted, including reducing structural complexity, adjusting process parameters, or recombining syntactic rules, until the generated result meets the application requirements. During or after the generation process, a multi-objective constraint optimization algorithm is introduced to jointly evaluate the matching degree of the generated result at the semantic, syntactic, and pragmatic levels, and optimize the selection of the generated result based on a weighted loss function.

[0185] The joint optimization objective function is expressed as:

[0186] L=λ1L sem +λ2L syn +λ3L pra ;

[0187] Among them, L sem L represents the loss of cultural semantic bias. syn L represents the loss due to structural process deviation. pra The loss represents the application scenario mismatch, with λ1, λ2, and λ3 being weighting coefficients. The verification content includes, but is not limited to, whether the process complexity exceeds a set threshold, whether the number of colors meets the weaving process limitations, and whether it conforms to the etiquette norms and usage requirements of the corresponding application scenario.

[0188] Once the verification is successful, a complete design scheme will be output. The design scheme will include at least the brocade pattern, color scheme description, corresponding weaving process parameters, and material list, which will be used to guide actual production or subsequent design applications.

[0189] This embodiment constructs a multimodal, multi-layered corpus for Nanjing brocade design, realizing the systematic expression of design source data. By introducing semantic, syntactic, and pragmatic layers, and combining multimodal data expression forms such as text, images, and structured parameters, it uniformly annotates and manages the cultural semantics, structural rules, and application scenarios related to Nanjing brocade design. This provides clear semantic basis, structural constraints, and pragmatic verification mechanisms for the design generation process, thereby improving design efficiency, enhancing the interpretability of the design process, and ensuring the cultural accuracy of the design results.

[0190] By simultaneously introducing semantic feature vectors, syntactic constraints, and pragmatic constraints into the design generation process, the generated scheme can be synergistically controlled in terms of cultural expression, structural rationality, and application feasibility, thus avoiding design deviations caused by relying solely on a single modality or a single level of control.

[0191] By using a multimodal corpus hierarchical annotation and feature mapping mechanism, the generated design schemes can clearly correspond to their semantic layer cultural semantics, grammatical layer structural rules, and pragmatic layer application scenario conditions. This improves the traceability, standardization, and controllability of the design results, making every design detail verifiable and facilitating the accumulation and reuse of design experience.

[0192] The design generation engine parses user-input design requirements into multi-layered design constraint parameters at the semantic, syntactic, and pragmatic levels. Based on this, it intelligently generates design solutions that meet the requirements of Nanjing brocade craftsmanship and application scenario limitations, achieving intelligent integration of the entire design process, significantly improving design efficiency, and reducing over-reliance on the experience of individual craftsmen.

[0193] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for designing and generating collaborative constraints for a multidimensional corpus of Yunjin brocade, characterized in that, Includes the following steps: S1. Construct a multi-dimensional hierarchical corpus of Nanjing brocade, standardize the multimodal data related to Nanjing brocade design, and perform hierarchical annotation according to semantic, syntactic and pragmatic layers. Set a unique corpus identifier for each Nanjing brocade corpus object, so that the same corpus object forms cross-layer bound hierarchical annotation data at different levels. S2. Based on the unique corpus identifier, establish cross-layer association rules between the semantic layer, syntactic layer and pragmatic layer to limit the legal combination paths between elements at different levels; S3. Extract semantic feature vectors, syntactic constraint parameters and pragmatic constraint parameters from the hierarchical labeled data respectively, and map the three layers of features to a unified feature space to form core element feature vectors. S4. Receive design requirements input by the user, perform hierarchical parsing of the design requirements, generate semantic feature vectors, syntactic constraint parameters and pragmatic constraint parameters, and perform legality screening of the constraint parameters based on the cross-layer association rules. S5. Input the selected multi-level collaborative constraint parameter set into the design generation engine to control the direction of cultural expression at the semantic level, control the composition topology and color organization at the grammatical level, and control the process complexity and production condition range at the pragmatic level to generate preliminary design results. S6. Perform feasibility verification on the preliminary design results based on pragmatic layer constraint parameters. If the preliminary design results do not meet the preset pragmatic layer constraint parameters, adjust the syntactic layer constraint parameters and re-execute the generation process. If the generation results simultaneously meet the consistency requirements of the semantic layer, syntactic layer, and pragmatic layer, output the design scheme.

2. The method for designing and generating collaborative constraints for a multidimensional corpus of Yunjin brocade as described in claim 1, characterized in that, The cross-layer association rule uses a three-dimensional rule matrix to represent the combination relationship between semantic layer elements, syntactic layer elements and pragmatic layer elements. When the matrix value meets the preset conditions, it means that the combination is allowed; otherwise, it means that the combination is prohibited. The legality of the combination path of the three-layer elements is screened before generation.

3. The method for designing and generating collaborative constraints for a multidimensional corpus of Yunjin brocade as described in claim 1, characterized in that, The logic for obtaining the semantic feature vector is as follows: The semantic feature vector is obtained by semantically encoding the text data using a large language model, and is represented as E. sem = f LLM (T), where f LLM (*) indicates a semantic encoding function based on a large language model; A semantic similarity algorithm is used to match the semantic feature vectors of design requirements with the semantic feature vectors of Yun brocade corpus objects. This is used to filter Yun brocade corpus objects that are consistent with the design requirements at the cultural and semantic level. The semantic similarity algorithm is as follows: ; where: E sem i E represents the i-th semantic feature vector; sem j Sim represents the j-th semantic feature vector; sem (i,j) represents the semantic similarity between the i-th semantic feature vector and the j-th semantic feature vector; ||·|| represents the magnitude of the vector.

4. The method for designing and generating collaborative constraints for a multidimensional corpus of Yunjin brocade as described in claim 3, characterized in that, The pattern image is subjected to structural topology recognition and parameterization calculation to obtain the parameters of the number of contour nodes, line density, and color layer distribution. After obtaining the semantic layer constraint set, matching syntactic rule entries are selected from the corpus according to the semantic layer constraint set. The selected syntactic rule entries are converted into syntactic layer constraint parameters, which are used to limit the composition topology, color layer organization, and weaving structure type during the design generation process.

5. The method for designing and generating collaborative constraints for a multidimensional corpus of Yunjin brocade as described in claim 4, characterized in that, The pragmatic layer constraint parameters include application scenario category, adaptation carrier type, production method parameters, and cost range parameters. A threshold mapping relationship is established between the grammatical layer structural complexity index and the production capacity range to verify the feasibility of the grammatical layer constraint parameters.

6. The method for designing and generating collaborative constraints for a multidimensional corpus of Yunjin brocade as described in claim 1, characterized in that, The logic for obtaining the feature vectors of the core elements includes: The semantic feature vector, syntactic constraint parameters and pragmatic constraint parameters are used to screen the legality of the three-layer feature combination through cross-layer association rules, and feature combinations that do not meet the cultural etiquette rules or structural and technological constraints are eliminated. The selected three-layer features are mapped to a unified feature space and fused according to preset weights to generate a core element feature vector. The core element feature vector is used to characterize the comprehensive expression state of the Yunjin corpus object in terms of cultural semantics, structural form and application attributes.

7. The method for designing and generating collaborative constraints for a multidimensional corpus of Yunjin brocade as described in claim 1, characterized in that, The construction of the multi-level collaborative constraint parameter set includes: The user-input design requirements are parsed into semantic feature vectors, syntactic constraint parameters, and pragmatic constraint parameters; the legality of the three constraint parameters is screened based on the cross-layer association rule matrix to form candidate constraint combinations; The feasibility of the syntactic layer constraint parameters is verified based on the pragmatic layer production condition range. When the syntactic layer constraint parameters are detected to exceed the production capacity range of the target application scenario, the syntactic layer parameter adjustment mechanism is triggered to shrink the structural complexity index. The constraint combination that has been screened and verified is determined as a multi-level collaborative constraint parameter set to control the design generation path.

8. The method for designing and generating collaborative constraints for a multidimensional corpus of Yunjin brocade as described in claim 1, characterized in that, The design generation engine employs a conditional generation model, which takes the fused semantic feature vector, syntactic constraint parameters, and pragmatic constraint parameters as conditional inputs. The generation process is guided by a conditional control mechanism, and its core generation process is as follows: G=f SD (E sem ,E syn ,E pra ); Among them: E sem E represents a semantic feature vector. syn E represents the constraint parameters of the grammatical structure level. pra f represents pragmatic layer constraint parameters; SD (E sem E syn E pra () indicates the conditional control mechanism used by the design generation engine; G indicates the guiding generation result; A low-rank adaptive parameter is introduced for fine-tuning, and the fine-tuning process satisfies: W' = W + ΔW, ΔW = AB Where W represents the original model parameters, and A∈R d×r B belongs to R r×k ΔW is a low-rank matrix, r≪min(d,k); ΔW is the low-rank weight increment matrix; W' is the fine-tuned model parameter, used to enable the model to learn the structural form and color organization rules of brocade patterns.

9. The method for designing and generating collaborative constraints for a multidimensional corpus of Yunjin brocade as described in claim 1, characterized in that, The joint optimization objective function introduced in the design generation process is expressed as: L=λ1L sem +λ2L syn +λ3L pra ? Among them, L sem L represents the loss of cultural semantic bias. syn L represents the loss due to structural process deviation. pra The loss represents the mismatch between the application scenario and the target function, where λ1, λ2, and λ3 are weighting coefficients. When the joint optimization objective function exceeds a preset threshold, the design generation engine is triggered to re-execute the generation process.

10. A design generation system based on collaborative constraints of a multidimensional corpus of Nanjing brocade, used to run the design generation method based on collaborative constraints of a multidimensional corpus of Nanjing brocade as described in any one of claims 1-9, characterized in that, include: The multidimensional hierarchical corpus construction module is used to collect and organize multimodal data related to Nanjing brocade design, and divide it into semantic, syntactic and pragmatic layers, and set a unique corpus identifier for each Nanjing brocade corpus object; The hierarchical annotation module is used to perform semantic annotation, syntactic annotation and pragmatic annotation on the Yunjin corpus objects respectively, and to establish cross-layer association rules through the unique corpus identifier; The core element feature extraction module is used to extract semantic feature vectors, syntactic constraint parameters and pragmatic constraint parameters respectively, and map the features to a unified core element feature vector. The requirements parsing module is used to parse the user-input design requirements into semantic feature vectors, syntactic constraint parameters, and pragmatic constraint parameters. The design generation engine is used to introduce the multi-level collaborative constraint parameter set into the design generation engine, perform multi-level constraint control on the generation process, and generate design results. The pragmatic verification module is used to verify the pragmatic feasibility of the generated results and to adjust the generated parameters when the preset pragmatic layer constraint parameters are not met. The multimodal design generation output module is used to output design schemes that meet the consistency requirements of the semantic, syntactic, and pragmatic layers.

Images