Automobile design demand analysis and visual feature generation method based on large model

Abstract

The invention belongs to the technical field of artificial intelligence image generation, and particularly relates to an automobile design demand analysis and visual feature generation method based on a large model. Comprising the following steps: step 1, constructing an automobile modeling design semantic ontology model; 2, designing an ontology-driven large language model reasoning mechanism; and step 3, generating a structured modeling feature keyword and visually outputting the structured modeling feature keyword. According to the invention, a set of complete automobile modeling design semantic ontology model is realized, and the model scientifically constructs a rigorous mapping relation between user demand dimensions (gender, age, usage scene, style preference and price interval) and modeling feature dimensions (automobile type, color, aspect ratio, front and rear overhang length, automobile window ratio and shoulder line height). And a concept system and a logic rule are clearly defined. And the technical bottlenecks of inaccurate user demand understanding, imperfect semantic mapping mechanism, high natural language interaction threshold and the like in the existing automobile modeling design system are overcome.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence image generation technology, specifically a method for analyzing automotive design requirements and generating visual features based on a large model. Background Technology

[0002] With the booming development of the new energy vehicle market and the increasingly diversified aesthetic demands of consumers, automotive styling has become one of the core elements of brand differentiation. In the traditional automotive styling design process, designers mainly rely on personal experience and intuition for creation. This process is not only lengthy and inefficient, but also makes it difficult to systematically capture and respond to the personalized and diverse needs of users. Especially when faced with natural language descriptions from non-professional users, designers often need to spend a lot of time interpreting vague and abstract user needs and transforming them into specific styling feature parameters. This inefficient communication method seriously restricts the design response speed and user satisfaction. Currently, the field of automotive styling design urgently needs an intelligent design method that can accurately understand user intentions and efficiently generate styling solutions that meet those needs.

[0003] Ontology, as a structured method of knowledge representation, provides a theoretical foundation for solving the aforementioned problems. By clearly defining concepts, attributes, and their relationships, ontology constructs hierarchical semantic networks, effectively organizing and expressing domain-specific professional knowledge. In the field of product design, existing research has attempted to apply ontology to general product design knowledge management; however, these studies mostly focus on product functional structure or manufacturing processes, lacking targeted modeling for the highly subjective and multi-dimensionally coupled design process of automobile styling. In particular, in the specific scenario of automobile styling design, there is currently no mature method for systematically constructing a semantic mapping relationship between user needs and styling features, achieving a precise conversion from "what people need" to "what the car looks like." Traditional design methods often rely on the designer's personal experience and subjective judgment, lacking a unified semantic description framework and computable mapping rules, making it difficult to standardize and intelligentize the design process.

[0004] The emergence of large language models has brought new opportunities to solving the problem of understanding user needs. With their powerful language understanding and generation capabilities, large language models can effectively analyze the implicit intentions in users' natural language. However, large language models alone lack domain expertise and structured expression capabilities, resulting in significant limitations when handling specialized automotive styling design tasks. While some research has attempted to apply large language models to general product recommendation systems, these applications mostly remain at the level of simple keyword extraction, failing to deeply construct a systematic mapping mechanism between user needs and styling features, and even less able to achieve accurate conversion from natural language to structured styling features. Especially in the highly specialized field of automotive styling design, existing technologies lack both a professional knowledge base for automotive styling features and a semantic parsing mechanism capable of coordinating user subjective expressions with professional design specifications.

[0005] Diffusion models, as the most advanced image generation technology currently available, offer the possibility of visualizing styling designs. However, diffusion models typically require precise text prompts as input to generate high-quality images, while ordinary users often struggle to provide structured descriptions that meet professional requirements. Existing automotive design systems based on diffusion models largely rely on preset options or input from professional designers, failing to directly process users' natural language expressions, resulting in high system interaction barriers and poor user-friendliness. Although some research has attempted to improve structural control capabilities by combining technologies such as ControlNet, these methods have still not solved the problem of automatically converting users' natural language into structured prompts, creating a significant gap in the user input stage of the entire design process.

[0006] In the process of achieving automatic conversion from user needs to styling features in automotive styling design based on ontology and large language models, the following key challenges still exist:

[0007] Constructing a semantic system for automotive styling design is difficult;

[0008] Automotive styling design involves a multi-dimensional coupling of user needs and styling features. Constructing a semantic description system that comprehensively covers automotive styling features while accurately expressing user needs is a fundamental challenge for achieving intelligent design. Current technologies lack specialized semantic modeling methods for automotive styling design, making it difficult to establish a systematic mapping relationship between user needs and styling features.

[0009] Natural language understanding lacks accuracy and professionalism;

[0010] Users' natural language expressions have problems such as ambiguity, vagueness, and structural incompleteness. Existing large language models often suffer from misunderstanding or information omission when dealing with automotive design terminology and complex semantic relationships. This leads to discrepancies between the extracted structured labels and the user's true intentions, affecting the accuracy of subsequent styling generation.

[0011] The mechanisms for handling missing information and semantic conflicts are inadequate;

[0012] In user natural language input, key dimensional information is often missing (such as age or gender not being explicitly mentioned) or semantic conflicts frequently occur. Existing systems lack effective reasoning completion and conflict resolution mechanisms, making it difficult to generate complete, consistent, and reasonable structured label sets, thus limiting the system's practicality and robustness.

[0013] The difficulty lies in coordinating ontological rules with the user's subjective intent;

[0014] Automotive design combines functionality and artistry, and there is a potential conflict between the mapping rules built on ontology models based on design experience and the personalized and subjective expressions of users. When user needs are inconsistent with the recommendations of ontology rules, existing systems struggle to strike a balance between adhering to professional knowledge and respecting user preferences, resulting in generated results that neither conform to professional standards nor meet user expectations. Summary of the Invention

[0015] This invention provides a method for analyzing automotive design requirements and generating visual features based on a large model. The core innovation lies in implementing a complete semantic ontology model for automotive styling design. This model scientifically constructs a rigorous mapping relationship between user requirement dimensions (gender, age, usage scenario, style preference, price range) and styling feature dimensions (model, color, aspect ratio, front and rear overhang length, window ratio, shoulder line height), clearly defining the conceptual system and logical rules. Based on this, the invention innovatively designs a large language model reasoning mechanism guided by ontology knowledge, effectively solving the "semantic gap" problem between user requirements and professional expressions in automotive styling design. It achieves accurate analysis and structured transformation of vague and abstract user natural language descriptions. This mechanism can automatically extract and complete key dimension information in user requirements without requiring users to understand professional tag systems, coordinate and handle semantic conflicts, and generate styling feature keywords that conform to both automotive design expertise and the user's true intentions. This significantly improves the intelligence level and user-friendliness of the design process, overcoming technical bottlenecks in existing automotive styling design systems such as inaccurate understanding of user requirements, imperfect semantic mapping mechanisms, and high barriers to natural language interaction.

[0016] The technical solution of this invention is described below in conjunction with the accompanying drawings:

[0017] This invention provides a method for analyzing automotive design requirements and generating visual features based on a large model, comprising the following steps:

[0018] Step 1: Construct a semantic ontology model for automotive styling design;

[0019] Step 2: Design an ontology-driven reasoning mechanism for a large language model;

[0020] Step 3: Generate structured model feature keywords and output them visually.

[0021] Furthermore, the specific method for step one is as follows:

[0022] 11) Defining the scope of the ontology domain and constructing the core concept system;

[0023] 12) Ontology hierarchy and relation modeling;

[0024] 13) Construction of a rule-based mapping knowledge base;

[0025] 14) The programmatic implementation of the ontology and its integration with the knowledge base.

[0026] Furthermore, the specific method for step 11) is as follows:

[0027] Through a systematic analysis of the automotive design field and combined with the results of a large-scale user questionnaire survey, the scope of the ontology is clearly defined as a binary relationship of "user needs - styling features". Then, two core concept classes are proposed: UserRequirement and DesignFeature. UserRequirement represents the set of user needs in the context of car purchase and use, while DesignFeature represents the set of professional design elements that constitute the exterior form of a car and can be quantified or classified. UserRequirement is divided into five subclasses: Gender, AgeGroup, UsageScenari, PreferenceStyle, and PriceRange. DesignFeature is divided into six subclasses: CarType, Color, WidthHeightRatio, Overhang, WindowRatio, and ShoulderLine. Through this classification system, a semantic knowledge framework is constructed.

[0028] The specific method for step 12) is as follows:

[0029] The Web Ontology Language (OWL) is used as the formal modeling language, and the Protégé open-source tool is used for visual construction and maintenance. In terms of class hierarchy, the ontology uses UserRequirement and DesignFeature as the top-level classes, and derives subclasses layer by layer. In terms of relation modeling, the key object property hasRecommendedFeature is defined to establish the mapping relationship between user requirements and design features. The domain of the property is UserRequirement and the value domain is DesignFeature. During instantiation, specific requirement instances and design instances are introduced into the ontology, and the two are linked through the hasRecommendedFeature property, thereby realizing a structured link at the knowledge level.

[0030] The specific method for step 13) is as follows:

[0031] A rule-based mapping knowledge base is established within the ontology framework. The knowledge base consists of logical rules used to represent statistical regularities. The rules adopt the "If-Then" structure of first-order predicate logic. The rule construction process includes three stages: first, collecting matching data between user needs and styling features; second, using the matching data to extract the need-feature mapping pattern; and finally, reviewing and revising the rules to ultimately form the core rules.

[0032] The specific method for step 14) is as follows:

[0033] The ontology model is transformed into an executable knowledge base, which involves three aspects: First, the completed ontology is exported as a standardized ".owl" file; second, the Apache Jena framework is used as the ontology mapping engine, and the provided APIs are used to complete the parsing, querying, and logical reasoning of the ontology; using the SPARQL query language, the system automatically retrieves the corresponding "hasRecommendedFeature" instance, i.e., the recommended combination of modeling features, given a user requirement instance; finally, the ontology file and the Jena framework are integrated into the system backend in the Eclipse development environment, so that the knowledge base provides an interface as an independent module, providing standardized knowledge support for the subsequent large language model reasoning mechanism.

[0034] Furthermore, the specific method for step two is as follows:

[0035] 21) Provide guidance on engineering and ontology knowledge;

[0036] 22) Implement an automatic completion mechanism for user request information;

[0037] 23) Semantic conflict detection and coordination;

[0038] 24) Tag fusion and consistency check.

[0039] Furthermore, the specific method for step 21) is as follows:

[0040] First, by designing a prompting engineering approach for the automotive styling design field, the ontology knowledge system is embedded into the input of the large language model, ChatGPT-4o, thereby guiding the model to follow the predetermined semantic boundaries during natural language parsing. The prompt template contains a complete ontology structure of user needs and styling features, as well as tag extraction instructions, output format constraints, and explanations of key technical terms.

[0041] The specific method for step 22) is as follows:

[0042] An ontology-based rule-based automatic completion mechanism is adopted; when missing dimensional information is detected, the system performs reasoning and completion based on contextual semantic analysis and association rules in the ontology; the automatic completion mechanism is logically formalized as formula (1):

[0043] u′=u∪{a i |a i ∈A,Inference(u,a i )} (1)

[0044] In the formula, u is the original set of demand tags; a i The complete part is inferred through ontology reasoning; u′ is the final complete set of user requirement tags;

[0045] The specific method for step 23) is as follows:

[0046] An ontology-based semantic conflict detection and coordination mechanism is adopted; the semantic conflict detection and coordination mechanism first identifies logical conflicts between requirement tags, and then generates multiple sets of styling feature schemes through parallel reasoning.

[0047] The specific method for step 24) is as follows:

[0048] A multi-source information fusion strategy is adopted to compare and verify the preliminary output results of the large language model, the ontology rule recommendation results, and the user's original input; the three-party information is integrated through weighted voting, and the label combination with the highest consistency is selected as the final output as shown in formula (2). ：

[0049]

[0050] In the formula, D i is the candidate shape feature combination; Sim is the similarity function; w1, w2, and w3 are weighting coefficients.

[0051] Furthermore, the specific method for step three is as follows:

[0052] 31) Generate structured modeling feature keywords;

[0053] 32) Visualization of solutions and verification of technical effects based on diffusion models;

[0054] 33) Implement a user feedback and iterative optimization mechanism.

[0055] Furthermore, the specific method for step 31) is as follows:

[0056] After the multi-label coordination and fusion module completes its processing, the system finally outputs a set of modeling feature keywords K = {d1, d2, ..., d...} m}, where d i ∈D, where D is the complete set of shape features defined by the ontology; the complete set of shape features contains the precise values ​​of six core shape features: CarType, Color, WidthHeightRatio, Overhang, WindowRatio, and ShoulderLine; the keyword generation process has the following academic characteristics: First, consistency guarantee; all feature values ​​are strictly derived from the predefined enumeration set of the ontology model; Second, traceability; each feature value is traced back to the generation path, i.e., which ontology rule triggered it, and the contribution weight of the large language model and user input; Third, formalized output; the keyword set is encapsulated into an intermediate representation that can be read in JSON or XML.

[0057] The specific method for step 32) is as follows:

[0058] The keyword set is converted into prompt words that the diffusion model can recognize, and then an image is generated. Subsequently, the constructed prompt words are input into the general diffusion model, and the generated image result is shown in formula (3):

[0059] I=G(K,θ) (3)

[0060] In the formula, G is the diffusion model; K is the keyword set; and θ is the diffusion model parameter.

[0061] The specific method for step 33) is as follows:

[0062] A user feedback-driven closed-loop optimization mechanism is adopted; users interact with the generated results, including selecting the most satisfactory solution, fine-tuning individual design features, or rejecting the entire solution and re-entering a description; user feedback is modeled as Feedback. t And through the update function K t+1 =F(K) t Feedback tThis is then integrated into the next round of reasoning to generate an optimized set of feature keywords and visualization results.

[0063] The beneficial effects of this invention are:

[0064] 1) First-ever construction of a professional ontology semantic description system in the field of automotive styling design: This invention innovatively constructs an ontology model specifically for the field of automotive styling design, and for the first time systematically defines the scientific mapping relationship between user demand dimensions (gender, age, usage scenario, style preference, price range) and styling feature dimensions (model, color, width-to-height ratio, front and rear overhang length, window ratio, shoulder line height); This ontology model fills the gap in systematic semantic modeling in the field of automotive styling design, and provides a structured and computable semantic support framework for the automatic conversion of user needs to styling features, significantly improving the standardization and intelligence level of the automotive styling design process;

[0065] 2) Innovative Ontology-Driven Large Language Model Reasoning Mechanism: This invention innovatively designs an ontology-guided large language model reasoning mechanism, effectively solving the "semantic gap" problem between users' natural language and professional styling features. This mechanism injects ontology structural knowledge into the large language model through prompting engineering, enabling the model to accurately understand the professional terminology and logical relationships in the field of automotive styling design. Without requiring users to understand the professional tag system, it automatically extracts and completes key dimension information from user needs, coordinates semantic conflicts, and generates styling feature keywords that conform to both automotive design expertise and the user's true intentions, significantly improving the accuracy and completeness of requirement analysis.

[0066] 3) Efficiently solves the problems of missing information and semantic conflicts: The automatic completion mechanism and conflict coordination algorithm proposed in this invention effectively solve the problems of missing key dimension information (such as not explicitly mentioning age or gender) and semantic conflicts (such as expressing both "family" and "sporty" style preferences) in users' natural language input; through the collaborative work of ontology rules and large language models, the system can generate different combinations of styling features for users to choose from, which significantly improves the practicality and robustness of the system and provides a more humanized and intelligent interactive experience for automotive styling design;

[0067] 4) Simplified Design Process and Lowered Professional Barrier: This invention enables non-professional users to directly express their design intentions using natural language. The system automatically converts these intentions into structured descriptions of styling features that professional designers can understand, completely changing the inefficient traditional "one-way creation, iterative modification" work method. This method significantly lowers the barrier to entry for automotive styling design, allowing ordinary users to easily participate in the design process. It promotes the intelligent transformation of automotive styling design from "designer-led" to "user-demand-driven," providing technical support for the automotive industry to achieve a truly user-centric design philosophy. Attached Figure Description

[0068] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0069] Figure 1 This is a flowchart of the present invention;

[0070] Figure 2 This is a schematic diagram of the ontology model structure;

[0071] Figure 3 This is a schematic diagram of the reasoning mechanism;

[0072] Figure 4 This is a schematic diagram of the system architecture. Detailed Implementation

[0073] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0074] See Figures 1-4 This invention provides a method for analyzing automotive design requirements and generating visual features based on a large model, comprising the following steps:

[0075] Step 1: Construct a semantic ontology model for automotive styling design. The specific method is as follows:

[0076] This step aims to establish a formalized, computable semantic knowledge framework for the field of automotive styling design, addressing the "semantic gap" between user needs and professional styling features. The ontology model, as the core knowledge source of the entire method, provides rigorous semantic constraints and a logical reasoning foundation for subsequent large language model inference, specifically:

[0077] 11) Defining the scope of the ontology domain and constructing the core concept system;

[0078] This invention focuses on the mapping of user's emotional needs to professional styling features. Through a systematic analysis of the automotive design field and combined with the results of a large-scale user questionnaire survey, the scope of the ontology is clearly defined as a binary relationship of "user needs - styling features". Based on this, two core concept classes are proposed: UserRequirement and DesignFeature. UserRequirement represents the set of subjective needs of users in the context of car purchase and use, while DesignFeature represents the set of quantifiable or categorizable professional design elements that constitute the exterior morphology of a car. To ensure the completeness and operability of the ontology model, the UserRequirement class is further subdivided into five subclasses: Gender, AgeGroup, UsageScenario, PreferenceStyle, and PriceRange; the DesignFeature class is subdivided into six subclasses: CarType, Color, WidthHeightRatio, Overhang, WindowRatio, and ShoulderLine. Through the aforementioned mutually exclusive and comprehensive classification system, a semantic knowledge framework with broad coverage and a reasonable structure is constructed.

[0079] 12) Ontology hierarchy and relation modeling;

[0080] To achieve semantic consistency and reasonability, this invention employs Web Ontology Language (OWL) as the formal modeling language and utilizes the Protégé open-source tool for visual construction and maintenance. Regarding the class hierarchy, the ontology uses UserRequirement and DesignFeature as top-level classes, deriving subclasses layer by layer. For example, PreferenceStyle can be further subdivided into Sports, Business, and Home. In terms of relationship modeling, this invention defines a key object attribute, hasRecommendedFeature, used to establish the mapping relationship between user requirements and design features. This attribute has the domain UserRequirement and the value domain DesignFeature, formally expressing the logical semantics that "a certain combination of user requirements corresponds to a certain combination of design features." Furthermore, during instantiation, based on user survey data, specific requirement instances and design instances are introduced into the ontology and linked through the hasRecommendedFeature attribute, thereby achieving structured links at the knowledge level.

[0081] 13) Construction of a rule-based mapping knowledge base;

[0082] To achieve automated reasoning from user needs to design features, this invention establishes a rule-based mapping knowledge base within an ontology framework. This knowledge base consists of a series of formalized logical rules used to represent domain expert experience and statistical regularities. The rules adopt an "If-Then" structure of first-order predicate logic, with the general form as follows:

[0083]

[0084] Where Condition(u) represents a conjunction composed of user requirement subclasses, for example (hasGender(u, Female) ∧ hasAgeGroup(u, 18-35) ∧ hasUsageScenario(u, Urban) ∧ hasPreferenceStyle(u, Sports) ∧ hasPriceRange(u, Medium)). The rule construction process includes three stages: First, through questionnaires, matching data of user requirements and style features are collected; second, high-confidence requirement-feature mapping patterns are extracted using the matching data; finally, the rules are reviewed and revised, ultimately forming 288 core rules. Taking a typical rule as an example, its logical expression is as follows:

[0085] If "Gender=Female AND AgeGroup=18-35AND UsageScenario=Urban ANDPreferenceStyle=Sports AND PriceRange=Medium";

[0086] Then "CarType=Sedan AND Color=White AND WidthHeightRatio=Low ANDOverhang=Short AND WindowRatio=High AND ShoulderLine=Low";

[0087] This rule base not only ensures the logical consistency of semantic mapping, but also provides computable knowledge support for subsequent reasoning.

[0088] 14) The procedural implementation of the ontology and its integration with the knowledge base;

[0089] To ensure the ontology model can be invoked and utilized in practical systems, this invention transforms it into an executable knowledge base. This involves three aspects: First, the completed ontology is exported as a standardized ".owl" file, achieving persistent knowledge storage and cross-platform compatibility. Second, the Apache Jena framework is used as the ontology mapping engine, leveraging its provided APIs to perform ontology parsing, querying, and logical reasoning. Using the SPARQL query language, the system can automatically retrieve corresponding "hasRecommendedFeature" instances (recommended combination of modeling features) based on a user's input requirement instance. Finally, the ontology file and the Jena framework are integrated into the system backend within the Eclipse development environment, allowing the knowledge base to provide an interface as an independent module, providing standardized knowledge support for subsequent large language model reasoning mechanisms. Through this implementation, the ontology model not only completes the transformation from the conceptual layer to the procedural layer but also decouples knowledge representation from application logic, ensuring the system's scalability and reusability. The completed ontology model is exported as an OWL format file, and the Apache Jena framework is used to implement procedural reading and manipulation of the ontology data. By using the Eclipse development environment, the ontology model and mapping rules are integrated into the system knowledge base, providing structured semantic support for the subsequent large language model reasoning mechanism and ensuring that ontology knowledge can be effectively accessed and utilized.

[0090] Step 2: Design an ontology-driven large language model reasoning mechanism. The specific method is as follows:

[0091] 21) Provide guidance on engineering and ontology knowledge;

[0092] In this invention, a prompting engineering approach is first designed for the automotive styling design field. This involves embedding the ontology knowledge system into the input of a large language model (ChatGPT-4o), guiding the model to adhere to predetermined semantic boundaries during natural language parsing. The prompt template not only contains a complete ontology structure of user needs and styling features but also includes tag extraction instructions, output format constraints, and explanations of key technical terms. Through this method, the model can automatically transform "fuzzy expressions" into standardized tags defined by the ontology when parsing natural language. For example, if a user inputs "stylish and dynamic small car," the large language model, guided by the prompt, will categorize "stylish and dynamic" as PreferenceStyle = Sports and "small car" as CarType = Sedan. The introduction of prompting engineering ensures the controllability and consistency of the semantic parsing process.

[0093] 22) Implement an automatic completion mechanism for user request information;

[0094] In practical applications, users' natural language input often lacks key information, such as not explicitly mentioning age or gender. To address this, this invention proposes an autocomplete mechanism based on ontology rules. When missing dimensional information is detected, the system performs inference and completion based on contextual semantic analysis and association rules in the ontology. For example, when a user inputs "cars suitable for young people commuting," the system infers AgeGroup = 18–35 based on "young people," while the gender dimension is reasonably inferred by combining style preferences and car model preferences. Logically, this mechanism can be formalized as formula (1):

[0095] u′=u∪{a i |a i ∈A,Inference(u,a i )} (1)

[0096] Where u is the original set of demand labels; a i The complete part is inferred through ontology reasoning; u′ is the final complete set of user requirement tags; this mechanism significantly improves the completeness and robustness of requirement parsing;

[0097] 23) Semantic conflict detection and coordination;

[0098] Users' natural language may contain conflicting needs, such as "wanting both sports and home use." To address this, this invention designs a semantic conflict detection and coordination mechanism based on ontology rules. This mechanism first identifies logical conflicts between need tags, and then generates multiple sets of design feature schemes through parallel reasoning, rather than simply fuzzily fusing the conflicting tags. For example, when a user inputs the description "wanting both home use and sports," the system decomposes it into two independent need sets, "home use" and "sports," and matches the corresponding design feature combinations to each. This mechanism avoids semantic inconsistencies caused by forced fusion, ensures the diversity and interpretability of the results, and provides users with flexible choices.

[0099] 24) Label fusion and consistency check;

[0100] After completing the requirement tag extraction, auto-completion, and conflict handling, the system needs to generate the final structured model feature set. This invention proposes a multi-source information fusion strategy, which compares and verifies the preliminary output of the large language model, the ontology rule recommendation results, and the user's original input. By combining the information from the three parties through weighted voting, the tag combination with the highest consistency is selected as the final output as shown in formula (2):

[0101]

[0102] In the formula, D iThe system uses candidate styling feature combinations; Sim is the similarity function; and w1, w2, and w3 are weighting coefficients. Through this fusion and verification mechanism, the system can ensure that the generated results not only conform to the professional knowledge of automotive styling design but also faithfully reflect the user's true intentions.

[0103] After extracting user demand tags and morphological features, a tag coordination and fusion mechanism is designed to compare and verify the preliminary results generated by the large language model with the ontology mapping rule base and the user's original input. This ensures that the generated results are both consistent with professional knowledge and aligned with the user's true intent, thus solving the problem of coordinating ontology mapping results with the user's subjective intent. This module employs a weighted voting strategy, combining information from the user's original input, ontology rule recommendations, and large language model inference to generate the final set of structured morphological feature keywords.

[0104] Step 3: Generate structured modeling feature keywords and visualize the output. The specific method is as follows:

[0105] 31) Generate structured modeling feature keywords;

[0106] After the multi-label coordination and fusion module completes its processing, the system finally outputs a highly standardized set of shape feature keywords K = {d1, d2, ..., d...} m}, where d i ∈D, where D is the complete set of styling features defined by the ontology. This set contains precise values ​​for six core styling features: CarType, Color, WidthHeightRatio, Overhang, WindowRatio, and ShoulderLine. The keyword generation process possesses the following academic characteristics: First, consistency guarantee. All feature values ​​are strictly derived from the predefined enumeration set of the ontology model (e.g., CarType∈{Sedan,SUV,MPV,SportsCar}), ensuring the standardization and unambiguity of the output. Second, traceability. Each feature value can be traced back to its generation path, i.e., which ontology rule (hasRecommendedFeature) triggered it, and the contribution weights of the large language model and user input, thereby improving the interpretability of the results. Third, formalized output. This keyword set is encapsulated into a machine-readable intermediate representation such as JSON or XML, ensuring compatibility with external applications (such as CAD systems, configurators, or diffusion models) and providing an interface for the modular expansion of the system.

[0107] 32) Visualization of solutions and verification of technical effects based on diffusion models;

[0108] To verify the effectiveness of the cross-modal chain from "natural language input—structured feature parsing—visual model generation", this step converts the keyword set into prompt words that the diffusion model can recognize and then generates images. Subsequently, the constructed prompt words are input into the general diffusion model, and the generated image results are shown in formula (3):

[0109] I=G(K,θ) (3)

[0110] In the formula, G represents the diffusion model; K represents the keyword set; and θ represents the diffusion model parameters. Compared with traditional geometric modeling-based design methods, this process more realistically and intuitively reflects the consistency between semantic features and visual form. The generated image is not only used for display but also serves to verify the front-end parsing quality of this invention: if the keywords can be correctly restored to the visual form, it indicates that the accuracy and effectiveness of semantic parsing have been verified.

[0111] 33) Implement a user feedback and iterative optimization mechanism;

[0112] To further enhance the system's adaptability and user-centricity, this invention designs a user feedback-driven closed-loop optimization mechanism. Users can interact with the generated results, including selecting the most satisfactory solution, fine-tuning individual design features (such as changing the color from white to blue), or rejecting the entire solution and re-entering a description. User feedback is modeled as Feedback. t And through the update function K t+1 =F(K) t Feedback t This is then integrated into the next round of reasoning, generating an optimized set of feature keywords and a visualization result. This mechanism not only enhances user engagement and interactive experience but also embodies the idea of ​​proactive learning, enabling the system to continuously correct its internal mapping relationships in each iteration, gradually converging towards the user's true design intent. The academic significance lies in the fact that this invention elevates the entire method from a static "one-off converter" to a dynamic "human-machine collaborative design agent," demonstrating its adaptability, scalability, and user-centricity.

[0113] Example 2

[0114] In the demand analysis and styling feature generation for urban commuting scenarios, the user inputs a natural language description: "I am a 25-year-old woman looking for a small car suitable for urban commuting, with a stylish and dynamic appearance, and a budget of around 150,000 yuan." After receiving the user's natural language input, the system performs semantic analysis through an ontology-driven large language model inference mechanism. Based on a pre-set prompt template, the large language model accurately extracts user demand dimension information from the input text: gender is clearly identified as "female" through "female"; age is clearly identified as "18-35 years old" through "25 years old"; usage scenario is clearly identified as "urban" through "urban commuting"; style preference is identified as "sporty" through "stylish and dynamic"; and price range is identified as "medium" through "around 150,000 yuan" combined with the price range definition of the ontology model. The system inputs the extracted user demand tags into the automotive styling design semantic ontology model, triggering the corresponding rules in the ontology mapping rule base. The ontology model is based on 288 predefined mapping rules. When the rule "If {gender = female, age = 18-35 years old, usage scenario = city, style preference = sporty, price range = medium} is matched, then {car model = sedan, color = white, width-to-height ratio = low, front and rear overhangs = short, window ratio = high, shoulder line = low}" is recommended. The multi-tag coordination and fusion module verifies the consistency between the extracted results and the user input. After confirming no conflicts, the ontology recommendation result is directly adopted. The system ultimately generates the following structured styling feature keywords: User needs: female, 18-35 years old, city, sporty, medium; Styling features: sedan, white, low width-to-height ratio, short front and rear overhangs, high window ratio, low shoulder line. This embodiment demonstrates the accurate parsing capability of this invention when processing complete user need information. It does not require users to understand a professional tag system, and can accurately extract five-dimensional user need information directly from natural language. Through the ontology model, it achieves a scientific mapping with six-dimensional styling features, specifically solving the problem of accurately converting subjective descriptions such as "fashionable and dynamic" to the professional term "sporty".

[0115] Example 3

[0116] In the process of demand reasoning and completion in scenarios with missing information, a user inputs a natural language description: "I want a car suitable for family use, with ample space, a stable appearance, and a reasonable price." After receiving the user input, the system initiates an ontology-driven large language model reasoning mechanism. The large language model attempts to extract user demand dimension information but finds that gender and age dimensions are not explicitly mentioned, only identifying the usage scenario as "family use," the style preference as "business," and the price range as "medium." The system then initiates an automatic information completion mechanism. For the missing gender dimension, the system analyzes the contextual semantics and combines association rules in the ontology model for reasoning: since the user describes "suitable for family use" and emphasizes "ample space," the system infers, based on statistical patterns in the ontology knowledge base, that this demand is more likely to come from a female member of the main family, thus completing the gender dimension to "female." For the missing age dimension, the system, through the combination of "suitable for family use" and "reasonable price," and combining the age distribution characteristics of family car users in the ontology model, infers the age range to be "35-55 years old." After information completion, the system obtains complete user demand tags: female, 35-55 years old, family use, business use, medium. The ontology model matches these tags to corresponding rules: "If {gender = female, age = 35-55 years old, usage scenario = family use, style preference = business use, price range = medium}, then recommend {vehicle type = MPV, color = black, width-to-height ratio = large, front and rear overhangs = long, window ratio = large, shoulder line = high}". The system outputs structured styling feature keywords as follows: User demand: female, 35-55 years old, family use, business use, medium; Styling features: MPV, black, large width-to-height ratio, long front and rear overhangs, large window ratio, high shoulder line. This embodiment fully demonstrates the reasoning ability of this invention in handling information gaps. Through a large language model guided by ontology knowledge, it can intelligently infer missing dimensions, ensuring the completeness and rationality of the generated results, and solving the design deviation problem caused by incomplete user expression in traditional systems.

[0117] Example 4

[0118] In handling semantic conflict scenarios, a user inputs a natural language description: "I want a car that's suitable for family use but also has a sporty feel, offering both comfort and good handling, suitable for city commuting and frequent highway driving." After receiving the input, the system analyzes it using an ontology-driven large language model inference mechanism. It discovers that the user expresses multiple conflicting preferences on the same dimension: in the style preference dimension, it includes both "family use" and "sporty"; in the usage scenario dimension, it includes both "city" and "highway." To address this semantic conflict, the system initiates a semantic conflict handling mechanism. GPT first extracts parallel labels for the conflicting semantics, avoiding fuzzy fusion of different labels, and instead labeling them separately as multiple parallel labels. For the style preference dimension, the system generates two independent sets of labels: "family use" and "sporty"; for the usage scenario dimension, the system generates two independent sets of labels: "city" and "highway." Subsequently, the system combines these parallel tags into multiple sets of user requirement prompts, which are then input into the ontology inference engine: the first set is {female, 18-35 years old, city, home, medium}, the second set is {female, 18-35 years old, city, sports, medium}, the third set is {female, 18-35 years old, high speed, home, medium}, and the fourth set is {female, 18-35 years old, high speed, sports, medium}. Based on these four different sets of user demand tags, the ontology model matches them with corresponding mapping rules to generate four sets of styling feature suggestions: The first set is {Model = MPV, Color = Silver, Width-to-Height Ratio = Large, Front and Rear Overhangs = Long, Window Ratio = Large, Shoulder Line = Low}; the second set is {Model = Sedan, Color = Red, Width-to-Height Ratio = Small, Front and Rear Overhangs = Short, Window Ratio = Small, Shoulder Line = High}; the third set is {Model = SUV, Color = Black, Width-to-Height Ratio = Large, Front and Rear Overhangs = Long, Window Ratio = Medium, Shoulder Line = Medium}; the fourth set is {Model = Sports Car, Color = Blue, Width-to-Height Ratio = Small, Front and Rear Overhangs = Short, Window Ratio = Small, Shoulder Line = High}. The system provides these four sets of suggestions to the user for selection and modification. The user can choose the most satisfactory scheme according to personal preference, or combine and adjust multiple schemes. This embodiment demonstrates the coordination capability of the present invention in handling complex preferences expressed by users. Through a parallel reasoning mechanism, the system can effectively address the label conflict problem caused by multiple preference expressions while ensuring the logicality of the ontology reasoning structure, providing users with a diverse selection space and solving the limitation of traditional systems that can only provide a single compromise solution when facing semantic conflicts.

[0119] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing automotive design requirements and generating visual features based on a large model, characterized in that, Includes the following steps: Step 1: Construct a semantic ontology model for automotive styling design; Step 2: Design an ontology-driven reasoning mechanism for a large language model; Step 3: Generate structured model feature keywords and output them visually.

2. The method for analyzing automotive design requirements and generating visual features based on a large model according to claim 1, characterized in that, The specific method for step one is as follows: 11) Defining the scope of the ontology domain and constructing the core concept system; 12) Ontology hierarchy and relation modeling; 13) Construction of a rule-based mapping knowledge base; 14) The programmatic implementation of the ontology and its integration with the knowledge base.

3. The method for analyzing automotive design requirements and generating visual features based on a large model according to claim 2, characterized in that, The specific method for step 11) is as follows: Through a systematic analysis of the automotive design field and combined with the results of a large-scale user questionnaire survey, the scope of the ontology is clearly defined as a binary relationship of "user needs - styling features". Then, two core concept classes are proposed: UserRequirement and DesignFeature. UserRequirement represents the set of user needs in the context of car purchase and use, while DesignFeature represents the set of professional design elements that constitute the exterior form of a car and can be quantified or categorized. UserRequirement is divided into five subclasses: Gender, AgeGroup, UsageScenario, PreferenceStyle, and PriceRange. DesignFeature is divided into six subclasses: CarType, Color, WidthHeightRatio, Overhang, WindowRatio, and ShoulderLine. Through this classification system, a semantic knowledge framework is constructed. The specific method for step 12) is as follows: The Web Ontology Language (OWL) is used as the formal modeling language, and the Protégé open-source tool is used for visual construction and maintenance. In terms of class hierarchy, the ontology uses UserRequirement and DesignFeature as the top-level classes, and derives subclasses layer by layer. In terms of relation modeling, the key object property hasRecommendedFeature is defined to establish the mapping relationship between user requirements and design features. The domain of the property is UserRequirement and the value domain is DesignFeature. During instantiation, specific requirement instances and design instances are introduced into the ontology, and the two are linked through the hasRecommendedFeature property, thereby realizing a structured link at the knowledge level. The specific method for step 13) is as follows: A rule-based mapping knowledge base is established within the ontology framework. The knowledge base consists of logical rules used to represent statistical regularities. The rules adopt the "If-Then" structure of first-order predicate logic. The rule construction process includes three stages: first, collecting pairing data of user needs and styling features; second, using the pairing data to extract the need-feature mapping pattern; and finally, reviewing and revising the rules to form the core rules. The specific method for step 14) is as follows: The ontology model is transformed into an executable knowledge base, which involves three aspects: First, the completed ontology is exported as a standardized ".owl" file; second, the Apache Jena framework is used as the ontology mapping engine, and the provided APIs are used to complete the parsing, querying, and logical reasoning of the ontology; using the SPARQL query language, the system automatically retrieves the corresponding "hasRecommendedFeature" instance, i.e., the recommended combination of modeling features, given a user requirement instance; finally, the ontology file and the Jena framework are integrated into the system backend in the Eclipse development environment, so that the knowledge base provides an interface as an independent module, providing standardized knowledge support for the subsequent large language model reasoning mechanism.

4. The method for analyzing automotive design requirements and generating visual features based on a large model according to claim 1, characterized in that, The specific method for step two is as follows: 21) Provide guidance on engineering and ontology knowledge; 22) Implement an automatic completion mechanism for user request information; 23) Semantic conflict detection and coordination; 24) Tag fusion and consistency check.

5. The method for analyzing automotive design requirements and generating visual features based on a large model according to claim 4, characterized in that, The specific method for step 21) is as follows: First, by designing a prompting engineering approach for the automotive styling design field, the ontology knowledge system is embedded into the input of the large language model, ChatGPT-4o, thereby guiding the model to follow the predetermined semantic boundaries during natural language parsing. The prompt template contains a complete ontology structure of user needs and styling features, as well as tag extraction instructions, output format constraints, and explanations of key technical terms. The specific method for step 22) is as follows: An ontology-based rule-based automatic completion mechanism is adopted; when missing dimensional information is detected, the system performs reasoning and completion based on contextual semantic analysis and association rules in the ontology; the automatic completion mechanism is logically formalized as formula (1): u′=u∪{a i ∣a i ∈A,Inference(u,a i )} (1) In the formula, u is the original set of demand tags; a i The complete part is inferred through ontology reasoning; u′ is the final complete set of user requirement tags; The specific method for step 23) is as follows: An ontology-based semantic conflict detection and coordination mechanism is adopted; the semantic conflict detection and coordination mechanism first identifies logical conflicts between requirement tags, and then generates multiple sets of styling feature schemes through parallel reasoning. The specific method for step 24) is as follows: A multi-source information fusion strategy is adopted to compare and verify the preliminary output results of the large language model, the ontology rule recommendation results, and the user's original input; the three-party information is integrated through weighted voting, and the label combination with the highest consistency is selected as the final output as shown in formula (2). ： In the formula, D i is the candidate shape feature combination; Sim is the similarity function; w1, w2, and w3 are weighting coefficients.

6. The method for analyzing automotive design requirements and generating visual features based on a large model according to claim 1, characterized in that, The specific method for step three is as follows: 31) Generate structured modeling feature keywords; 32) Visualization of solutions and verification of technical effects based on diffusion models; 33) Implement a user feedback and iterative optimization mechanism.

7. The method for analyzing automotive design requirements and generating visual features based on a large model according to claim 6, characterized in that, The specific method for step 31) is as follows: After the multi-label coordination and fusion module completes its processing, the system finally outputs a set of modeling feature keywords K = {d1, d2, ..., d...} m }, where d i ∈D, where D is the complete set of shape features defined by the ontology; the complete set of shape features contains the precise values ​​of six core shape features: CarType, Color, WidthHeightRatio, Overhang, WindowRatio, and ShoulderLine; the keyword generation process has the following academic characteristics: First, consistency guarantee; all feature values ​​are strictly derived from the predefined enumeration set of the ontology model; Second, traceability; each feature value is traced back to the generation path, i.e., which ontology rule triggered it, and the contribution weight of the large language model and user input; Third, formalized output; the keyword set is encapsulated into an intermediate representation that can be read in JSON or XML. The specific method for step 32) is as follows: The keyword set is converted into prompt words that the diffusion model can recognize, and then an image is generated. Subsequently, the constructed prompt words are input into the general diffusion model, and the generated image result is shown in formula (3): I=G(K,θ) (3) In the formula, G is the diffusion model; K is the keyword set; and θ is the diffusion model parameter. The specific method for step 33) is as follows: A closed-loop optimization mechanism driven by user feedback is adopted; users can interact with the generated results, including selecting the most satisfactory solution, fine-tuning individual design features, or rejecting the entire solution and re-entering a description. User feedback is modeled as Feedback t And through the update function K t+1 =F(K t Feedback t This is then integrated into the next round of reasoning to generate an optimized set of feature keywords and visualization results.

Images