Automobile AI drawing method and system based on fusion competition strategy

Abstract

The invention relates to an automobile AI drawing method and system based on a fusion competition strategy, and belongs to the field of artificial intelligence. The method comprises the following steps: inputting a fusion competition strategy, and carrying out strategy decomposition and parameter quantization to obtain a strategy parameter table; performing strategy coding based on the strategy parameter table, and outputting a strategy vector; performing design feature mapping based on the strategy vector to obtain a design feature vector; performing feature layering based on the design feature vector, and outputting layering condition embedding through layering coding; embedding and constructing a diffusion model based on a layering condition, and preliminarily generating an automobile design image; and performing strategy consistency evaluation based on the automobile design image, performing mapping matrix adaptive optimization based on an evaluation result, and outputting a final automobile design drawing. According to the method, multiple elements such as market competition strategies, brand features and cost limitation described by natural languages are quantitatively fused into the AI drawing process, so that the scientificity, efficiency and market adaptability of automobile design are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence, and particularly relates to an automobile AI drawing method and system based on a fusion competition strategy. BACKGROUND

[0002] With the intensification of global automobile industry competition and the increasing diversification of consumer demand, automobile design has changed from a traditional artistic creation activity to a complex system engineering that needs to take into account market positioning, competition strategy, cost control and brand inheritance. The traditional automobile design process usually takes 6-12 months and involves the collaborative work of dozens of designers and engineers, which is costly and highly dependent on the personal experience and intuition of designers, making it difficult to accurately respond to market changes. Although CAD / CAID computer-aided design tools have emerged in recent years, these tools only improve design efficiency and do not fundamentally change the design paradigm. With the development of deep learning technology, generative AI has begun to be applied to the field of product design, but existing technologies are still limited to single links such as sketch generation or 3D modeling, lacking comprehensive consideration of business strategies. In particular, existing AI design tools cannot effectively convert marketing strategies (such as target user portraits, price positioning, and product differentiation) into specific design parameters, resulting in design solutions that often deviate from business goals and require extensive manual adjustments. At the same time, existing technologies are difficult to balance innovation and brand DNA inheritance, easily leading to a lack of brand recognition in design results. In addition, most existing AI design systems use open-loop architectures and lack self-optimization capabilities based on evaluation feedback, making it difficult to learn from historical success / failure cases. In the context of the rapid popularization of new energy vehicles and the growing demand for individualization among consumers, traditional design methods have been unable to meet the multiple needs of enterprises to shorten product development cycles, accurately target niche markets, maintain brand characteristics and control research and development costs. SUMMARY

[0003] To solve the above problems existing in the prior art, the application provides an automobile AI drawing method and system based on a fusion competition strategy.

[0004] The purpose of the application can be achieved by the following technical solutions: An automobile AI drawing method based on a fusion competition strategy, the implementation of the automobile AI drawing method based on a fusion competition strategy includes the following steps: Step S1: input the fusion competition strategy, perform strategy decomposition and parameter quantization, and obtain a strategy parameter table; Step S2: perform strategy encoding based on the strategy parameter table and output a strategy vector; Step S3: perform design feature mapping based on the strategy vector and obtain a design feature vector; Step S4: feature layering based on the design feature vector, and outputting a layered conditional embedding through layered coding; Step S5: constructing a diffusion model based on the layered conditional embedding, and preliminarily generating a car design image; Step S6: performing strategy consistency evaluation based on the car design image, and performing adaptive optimization of a mapping matrix based on the evaluation result, and outputting a final car design image.

[0005] Preferably, the strategy decomposition and parameter quantification in step S1 are specifically: describing the fusion competition strategy in natural language; decomposing the fusion competition strategy into sub-strategies, the sub-strategies including vehicle type information, target crowd, design style, competition requirements, and cost constraints; quantifying the fusion competition strategy based on the sub-strategies to obtain quantized values; generating the strategy parameter table based on the sub-strategies and the quantized values.

[0006] Preferably, the strategy coding in step S2 is specifically: extracting keywords from the strategy parameter table to obtain strategy keywords and design feature keywords; converting the strategy keywords and the design feature keywords into multi-dimensional word vectors; generating the strategy vector based on the multi-dimensional word vectors, which is mathematically described as wherein, is the strategy vector, is the i-th strategy weight, is the j-th design feature weight, is the word vector conversion function, is the i-th strategy keyword, is the j-th design feature keyword, n is the number of strategy keywords, and m is the number of design feature keywords, is the L2 norm.

[0007] Preferably, the design feature mapping in step S3 is specifically: presetting a mapping matrix, each row of the mapping matrix corresponding to a strategy keyword, each column corresponding to a design feature keyword, and each grid value representing the influence degree of the strategy on the design feature; presetting a brand DNA bias vector; obtaining the design feature vector through the mapping matrix and the brand DNA bias vector, which is mathematically described as wherein, is the design feature vector, is the mapping matrix, For the policy vector, This is the brand DNA bias vector.

[0008] Preferably, the hierarchical coding in step S4 specifically includes: The design feature vector is layered to obtain high-level features, mid-level features, and low-level features; The hierarchical encoding is performed using a hierarchical conditional injection formula, mathematically described as follows: ,in, For the conditional embedding of the u-th layer, It is a multilayer perceptron. To design the u-th layer slice of the feature vector, u=1 is the high-level feature, u=2 is the middle-level feature, and u=3 is the low-level feature; The hierarchical conditional embedding C={ , , }

[0009] Preferably, the construction of the diffusion model in step S5 specifically involves: Obtain the current noise image, and then... The hierarchical conditional embedding C and the current step number t are input into the noise prediction network to obtain the predicted noise image; Obtain the current noise image The similarity with the design feature vector is calculated, and the similarity is obtained for the current noisy image. Gradients, constructing policy guidance terms ,in, Let z be the standard deviation of the noise, and z be the random noise. The strategy guidance strength coefficient. For similarity to the current noisy image gradient, For the current noisy image With design feature vectors Similarity; Based on the predicted noise image and the policy guidance term, a policy-guided sampling formula is constructed, mathematically described as follows: ,in, For the noisy image at step t-1, For single-step noise scheduling parameters, For cumulative noise scheduling parameters, For noise prediction networks; When the noise image at step t-1 meets the preset conditions, the diffusion stops and the noise image at step t-1 is output as the car design image.

[0010] Preferably, the policy consistency evaluation in step S6 is specifically: The design feature values are extracted from the initially generated automobile design image, and a comprehensive score is obtained through a policy consistency comprehensive scoring formula, which is mathematically described as , wherein, is the comprehensive score, is the design feature value weight, is the i-th design feature value extracted from the automobile design image, is the i-th target design feature value, and N is the number of design feature values, is the competitive differentiation weight, KL is the KL divergence, is the feature distribution of the automobile design image, is the feature distribution of the competitor, is the indicator function, which returns 1 when the condition is true, and 0 otherwise, is the j-th cost-related feature value of the automobile design image, is the maximum allowed value of the j-th cost constraint, and M is the number of design feature values with costs; The policy consistency evaluation is performed based on the comprehensive score, and when the comprehensive score is lower than a preset threshold, the mapping matrix adaptive optimization is triggered.

[0011] Preferably, the mapping matrix adaptive optimization in step S6 is specifically: A preset mask matrix is used to perform the mapping matrix adaptive optimization, which is mathematically described as , wherein, is the mapping matrix of the k+1 iteration, is the mapping matrix of the k iteration, is the learning rate, is the design feature vector extracted from the automobile design image, is the target design feature vector, is the transpose of the strategy vector, is the element-wise multiplication, is the mask matrix.

[0012] An automobile AI image generation system based on a fusion competition strategy, which is used to perform the automobile AI image generation method based on the fusion competition strategy described above, includes a design feature mapping module, a hierarchical coding module, a diffusion model construction module, and an optimization module. The design feature mapping module is used to input the fusion competition strategy, perform strategy decomposition and parameter quantization, and obtain a strategy parameter table; based on the strategy parameter table, strategy coding is performed to output a strategy vector; based on the strategy vector, design feature mapping is performed to obtain a design feature vector. The hierarchical encoding module is used to perform feature hierarchical layering based on the design feature vector, and output hierarchical conditional embedding through hierarchical encoding. The diffusion model construction module is used to construct a diffusion model based on the hierarchical conditional embedding, and to initially generate a car design image; The optimization module is used to perform strategy consistency evaluation based on the car design image, and to perform adaptive optimization of the mapping matrix based on the evaluation results, and output the final car design drawing.

[0013] The beneficial effects of this invention are as follows: (1) Through strategy decomposition, parameter quantification and feature mapping, the precise transformation of business competition strategy into specific design elements is realized, so that the design results closely follow the market positioning and significantly improve the design success rate.

[0014] (2) By using a hierarchical feature encoding mechanism, conditional embeddings of corresponding levels are injected into different stages of the diffusion model to achieve full-dimensional control from the overall outline and main components to surface details, ensuring that the design scheme meets strategic requirements while maintaining aesthetic appeal.

[0015] (3) By using the brand DNA bias vector, we ensure that the generated design retains the core brand identification elements, solve the industry pain point that AI design can easily lead to the loss of brand characteristics, and maintain the brand asset value.

[0016] (4) By using strategy consistency assessment and mapping matrix adaptive optimization mechanism, a closed loop is formed, enabling the system to learn from historical cases autonomously, continuously improve design quality, and reduce manual intervention. Attached Figure Description

[0017] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0018] Figure 1 This is a flowchart of the steps of an AI-based image generation method for automobiles based on a fusion competition strategy according to the present invention. Detailed Implementation

[0019] To better understand the invention, various aspects of the invention will be described in more detail with reference to the accompanying drawings. It should be understood that these detailed descriptions are merely illustrative of exemplary embodiments of the invention and are not intended to limit the scope of the invention in any way. Throughout the specification, the expression "and / or" includes any and all combinations of one or more of the associated listed items. As used herein, the terms "approximately," "about," and similar terms are used as expressions of approximation, not as expressions of degree, and are intended to describe inherent deviations in measured or calculated values ​​that will be recognized by those skilled in the art. Furthermore, the order in which the steps are described in this invention does not necessarily indicate the order in which these steps occur in actual operation, unless otherwise expressly defined or deduced from the context.

[0020] It should also be understood that expressions such as "comprising," "including," "having," "containing," and / or "comprising" are open-ended rather than closed-ended expressions in this specification, indicating the presence of the stated features, elements, and / or components, but not excluding the presence of one or more other features, elements, components, and / or combinations thereof. Furthermore, when expressions such as "at least one of..." appear after a list of listed features, they modify the entire list of features, not just individual elements in the list. Additionally, when describing embodiments of the invention, the word "may" is used to mean "one or more embodiments of the invention." And the term "exemplary" is intended to refer to examples or illustrations.

[0021] Unless otherwise specified, all terms used herein (including engineering and technical terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that, unless expressly stated herein, terms defined in common dictionaries shall be interpreted as having the meaning consistent with their meaning in the context of the relevant art, and not in an idealized or overly formalized sense.

[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other. The invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0023] Example 1: Please see Figure 1 A method for generating automotive AI images based on a fusion competition strategy, comprising: Step S1: Input the fusion competition strategy, perform strategy decomposition and parameter quantization to obtain the strategy parameter table; Step S2: Encode the policy based on the policy parameter table and output the policy vector; Step S3: Perform design feature mapping based on the strategy vector to obtain the design feature vector; Step S4: Perform feature layering based on the designed feature vectors, and output layered conditional embeddings through layered encoding; Step S5: Construct a diffusion model based on the hierarchical conditional embedding to initially generate a car design image. The basic principle of the diffusion model is: starting from pure noise, removing noise step by step, and finally obtaining a clear image. Step S6: Perform strategy consistency evaluation based on the car design image, and perform adaptive optimization of the mapping matrix based on the evaluation results to output the final car design drawing.

[0024] In this embodiment, the strategy decomposition and parameter quantization specifically refer to: S101: The designer described the integrated competitive strategy in natural language, for example: We need a pure electric SUV for young urban families, emphasizing a balance between technology and practicality, with a price 20% lower than a certain car, and a distinctive appearance. S102: Decompose the fusion competition strategy to obtain sub-strategies. The sub-strategies include, but are not limited to, vehicle information (pure electric SUV), target audience (urban young families), design style (strong sense of technology and strong practicality), competitive requirements (price is 20% lower than a certain car, but it must be distinctive), and cost constraints (cannot use too expensive materials and processes). S103: Based on the sub-strategy, the parameters of the fusion competition strategy are quantified to obtain quantified values, such as: technological sense intensity: 0.8 (very strong); practicality intensity: 0.8 (very strong); cost control: 0.7 (relatively strong); differentiation demand: 0.6 (moderately strong). S104: Generate a structured strategy parameter table based on the sub-strategy and the quantized value.

[0025] In this embodiment, the policy encoding is specifically as follows: S201: Extract keywords from the strategy parameter table, extracting strategy keywords (technological feel, practicality, cost control, differentiation, etc.) and design feature keywords (through-type light strip, large space, economical materials, unique shape, etc., keywords that can satisfy the strategy). S202: Perform word vector conversion on the strategy keywords and the design feature keywords to obtain a multi-dimensional word vector for each keyword (the dimensions must be consistent). For example, "sense of technology" can be converted into a 300-dimensional vector of [0.12, -0.34, 0.56, ..., 0.78]. S203: Generate the strategy vector based on the multi-dimensional word vector, mathematically described as follows: ,in, For the policy vector, Let i be the weight of the i-th policy. For the j-th design feature weight, This is a word vector conversion function. For the i-th strategy keyword, Let j be the j-th design feature keyword, n be the number of strategy keywords, and m be the number of design feature keywords. It is an L2 norm.

[0026] In this embodiment, the design feature mapping specifically refers to: S301: A preset mapping matrix, where each row corresponds to a strategy keyword (technological feel, practicality, cost control, differentiation, etc.), and each column corresponds to a design feature keyword (wheelbase, headlight style, wheel size, etc.). Each cell's value represents the degree of influence of the strategy on the design feature. Example:

[0027] S302: Preset brand DNA bias vector. Each car brand has its own design style, such as the double kidney grille of brand A and the large mouth grille of brand B. The brand DNA bias vector can ensure that the final output has the brand characteristics. S303: The design feature vector is obtained through the mapping matrix and the brand DNA bias vector, mathematically described as follows: ,in, The design feature vector is m-dimensional, with each dimension representing the strength of a design feature; typically, 500 dimensions are used. Let be the mapping matrix (m×n). The policy vector is n-dimensional. This is the brand DNA bias vector (m-dimensional).

[0028] In this embodiment, the hierarchical coding specifically refers to: S401: Generally, AI rendering is done in steps and layers. First, the overall outline is generated, then the main components are added, and finally the details are added. Therefore, the design feature vector needs to be layered to obtain high-level features (features related to the overall shape, such as vehicle length, vehicle height, wheelbase, surface style, etc., generally 1-100 dimensions), mid-level features (features related to the main components, such as front face design, headlight style, window lines, etc., generally 101-300 dimensions), and low-level features (features related to detail processing, such as surface texture, seam width, chamfer size, etc., generally 301-500 dimensions). S402: The hierarchical encoding is performed through a hierarchical conditional injection formula, mathematically described as follows: ,in, For the conditional embedding of the u-th layer, It is a multilayer perceptron. To design the u-th layer slice of the feature vector, u=1 is the high-level feature, u=2 is the middle-level feature, and u=3 is the low-level feature; S403: Obtain the hierarchical conditional embedding C={ , , }

[0029] In this embodiment, the diffusion model is constructed as follows: S501: Obtain the current noisy image (initially a pure noise image), and then... The hierarchical conditional embedding C and the current step number t are input into the noise prediction network to obtain the predicted noise image (i.e., which pixels should be removed). The conditional embeddings of different levels play a role in different stages. Generally, the diffusion process is about 50 steps. In the early steps (t close to 50), high-level features are mainly used. In the middle steps (t is between 40 and 20), mid-level features are mainly used. In the later steps (t is less than 20), low-level features are mainly used. S502: Obtain the current noise image The similarity with the design feature vector is calculated, and the similarity is obtained for the current noisy image. Gradients, constructing policy guidance terms ,in, Let z be the noise standard deviation, and z be random noise. Same dimension, This is the strategy guidance strength coefficient (typical value 0.1-0.3). For similarity to the current noisy image gradient, For the current noisy image With design feature vectors The similarity factor is calculated by the guide term, which adjusts the noise added during the sampling process by calculating the similarity gradient between the currently generated image and the target design features, so that the generation process is directed in the direction of improving similarity. S503: Construct a policy-guided sampling formula based on the predicted noise image and the policy guidance term, mathematically described as follows: ,in, This is the noisy image at step t-1 (i.e., the image with less noise obtained in the next step, which is a reverse process). For single-step noise scheduling parameters (meaning that in the process of...) arrive In the noise reduction process, how much of the original signal is retained and how much noise is removed? This is the cumulative noise scheduling parameter (the sum of the single-step noise scheduling parameters from the initial pure noise image to the current step). For noise prediction networks; S504: When the noise image at step t-1 meets the preset conditions, stop the diffusion and output the noise image at step t-1 as the car design image.

[0030] In this embodiment, the policy consistency assessment specifically includes: S601: Extract design feature values ​​from the initially generated vehicle design image, and obtain a comprehensive score through the strategy consistency comprehensive scoring formula, mathematically described as follows: ,in, For comprehensive scoring, To design eigenvalue weights, Let i be the i-th design feature value extracted from the car design image (the image is analyzed and the design features are extracted using another AI model). Design eigenvalues ​​for the i-th target (from...) N is the number of design eigenvalues. For competitive differentiation weights, KL represents the KL divergence. Feature distribution for car design images For competitor feature distribution, This is an indicator function that returns 1 if the condition is true, and 0 otherwise. Let j be the cost-related feature value of the car design image. Let M be the maximum allowable value for the j-th cost constraint, and M be the number of design feature values ​​with existing costs. S602: Based on the comprehensive score, perform the strategy consistency evaluation. When the comprehensive score is lower than a preset threshold, trigger the adaptive optimization of the mapping matrix.

[0031] In this embodiment, the adaptive optimization of the mapping matrix specifically involves: S603: A preset mask matrix is ​​used to control which weights in the mapping matrix can be adjusted and by what extent. Adaptive optimization of the mapping matrix is ​​performed based on the mask matrix. Mathematically, this is described as follows: ,in, Let be the mapping matrix for the (k+1)th iteration. Let be the mapping matrix for the k-th iteration. For learning rate, This refers to the design feature vector extracted from automotive design images. Design feature vectors for the target (i.e.) ), This is the transpose of the policy vector. For element-wise multiplication, The mask matrix is ​​used; the iterated mapping matrix is ​​used to restart step S3 until the overall score of the generated image meets the standard, and finally the final car design drawing that meets the requirements is output.

[0032] Example 2: An AI-powered automotive image generation system based on a fusion competition strategy includes a feature mapping module, a hierarchical encoding module, a diffusion model construction module, and an optimization module. The design feature mapping module is used to input the fusion competition strategy, perform strategy decomposition and parameter quantization to obtain a strategy parameter table; perform strategy encoding based on the strategy parameter table to output a strategy vector; and perform design feature mapping based on the strategy vector to obtain a design feature vector. The hierarchical encoding module is used to perform feature hierarchical layering based on the design feature vector, and output hierarchical conditional embedding through hierarchical encoding. The diffusion model construction module is used to construct a diffusion model based on the hierarchical conditional embedding, and to initially generate a car design image; The optimization module is used to perform strategy consistency evaluation based on the car design image, and to perform adaptive optimization of the mapping matrix based on the evaluation results, and output the final car design drawing.

[0033] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for generating automotive AI images based on a fusion competition strategy, characterized in that, Includes the following steps: Step S1: Input the fusion competition strategy, perform strategy decomposition and parameter quantization to obtain the strategy parameter table; Step S2: Encode the policy based on the policy parameter table and output the policy vector; Step S3: Perform design feature mapping based on the strategy vector to obtain the design feature vector; Step S4: Perform feature layering based on the designed feature vectors, and output layered conditional embeddings through layered encoding; Step S5: Construct a diffusion model based on the hierarchical conditional embedding to initially generate a car design image; Step S6: Perform strategy consistency evaluation based on the car design image, and perform adaptive optimization of the mapping matrix based on the evaluation results to output the final car design drawing.

2. The automotive AI mapping method based on a fusion competition strategy according to claim 1, characterized in that, The strategy decomposition and parameter quantization in step S1 are specifically as follows: Describe the fusion competition strategy using natural language; The integrated competitive strategy is decomposed into sub-strategies, which include vehicle information, target audience, design style, competitive requirements, and cost constraints. Based on the sub-strategy, the parameters of the fusion competition strategy are quantized to obtain quantized values; The strategy parameter table is generated based on the sub-strategy and the quantized value.

3. The automotive AI mapping method based on a fusion competition strategy according to claim 1, characterized in that, The strategy encoding in step S2 is specifically as follows: Keyword extraction is performed on the strategy parameter table to extract strategy keywords and design feature keywords; The strategy keywords and design feature keywords are transformed into multi-dimensional word vectors. The strategy vector is generated based on the multidimensional word vectors, mathematically described as follows: ,in, For the policy vector, Let i be the weight of the i-th policy. For the j-th design feature weight, This is a word vector conversion function. For the i-th strategy keyword, Let j be the j-th design feature keyword, n be the number of strategy keywords, and m be the number of design feature keywords. It is an L2 norm.

4. The automotive AI mapping method based on a fusion competition strategy according to claim 1, characterized in that, The design feature mapping in step S3 specifically refers to: A preset mapping matrix is ​​provided, wherein each row of the mapping matrix corresponds to a strategy keyword, each column corresponds to a design feature keyword, and the value of each cell represents the degree of influence of the strategy on the design feature. Preset brand DNA bias vector; The design feature vector is obtained through the mapping matrix and the brand DNA bias vector, mathematically described as follows: ,in, To design feature vectors, For the mapping matrix, For the policy vector, This is the brand DNA bias vector.

5. The automotive AI mapping method based on a fusion competition strategy according to claim 1, characterized in that, The hierarchical coding in step S4 specifically refers to: The design feature vector is layered to obtain high-level features, mid-level features, and low-level features; The hierarchical encoding is performed using a hierarchical conditional injection formula, mathematically described as follows: ,in, For the conditional embedding of the u-th layer, It is a multilayer perceptron. To design the u-th layer slice of the feature vector, u=1 is the high-level feature, u=2 is the middle-level feature, and u=3 is the low-level feature; The hierarchical conditional embedding C={ , , } 6. The automotive AI mapping method based on a fusion competition strategy according to claim 1, characterized in that, The construction of the diffusion model in step S5 is specifically as follows: Obtain the current noise image, and then... The hierarchical conditional embedding C and the current step number t are input into the noise prediction network to obtain the predicted noise image; Obtain the current noise image The similarity with the design feature vector is calculated, and the similarity is obtained for the current noisy image. Gradients, constructing policy guidance terms ,in, Let z be the standard deviation of the noise, and z be the random noise. The strategy guidance strength coefficient. For similarity to the current noisy image gradient, For the current noisy image With design feature vectors Similarity; Based on the predicted noise image and the policy guidance term, a policy-guided sampling formula is constructed, mathematically described as follows: ,in, For the noisy image at step t-1, For single-step noise scheduling parameters, For cumulative noise scheduling parameters, For noise prediction networks; When the noise image at step t-1 meets the preset conditions, the diffusion stops and the noise image at step t-1 is output as the car design image.

7. The automotive AI mapping method based on a fusion competition strategy according to claim 1, characterized in that, The strategy consistency assessment in step S6 specifically involves: Design feature values ​​are extracted from the initially generated car design image, and a comprehensive score is obtained through the strategy consistency comprehensive scoring formula, mathematically described as follows: ,in, For comprehensive scoring, To design eigenvalue weights, Let i be the i-th design feature value extracted from the car design image. Design eigenvalues ​​for the i-th target, where N is the number of eigenvalues ​​to design. For competitive differentiation weights, KL represents the KL divergence. Feature distribution for car design images For competitor feature distribution, This is an indicator function that returns 1 if the condition is true, and 0 otherwise. Let j be the cost-related feature value of the car design image. Let M be the maximum allowable value for the j-th cost constraint, and M be the number of design feature values ​​with existing costs. The strategy consistency is evaluated based on the comprehensive score. When the comprehensive score is lower than a preset threshold, the mapping matrix is ​​adaptively optimized.

8. The automotive AI mapping method based on a fusion competition strategy according to claim 7, characterized in that, The adaptive optimization of the mapping matrix in step S6 specifically involves: A preset mask matrix is ​​used, and the mapping matrix is ​​adaptively optimized based on the mask matrix. Mathematically, this is described as follows: ,in, Let be the mapping matrix for the (k+1)th iteration. Let be the mapping matrix for the k-th iteration. For learning rate, This refers to the design feature vector extracted from automotive design images. Design feature vectors for the target. This is the transpose of the policy vector. For element-wise multiplication, It is a mask matrix.

9. A car AI mapping system based on a fusion competition strategy, characterized in that, The system is applied to the automotive AI mapping method based on a fusion competition strategy as described in any one of claims 1-8, and includes a feature mapping design module, a hierarchical encoding module, a diffusion model construction module, and an optimization module. The design feature mapping module is used to input the fusion competition strategy, perform strategy decomposition and parameter quantization, and obtain the strategy parameter table. Based on the policy parameter table, policy encoding is performed, and a policy vector is output. Based on the strategy vector, design feature mapping is performed to obtain the design feature vector; The hierarchical encoding module is used to perform feature hierarchical layering based on the design feature vector, and output hierarchical conditional embedding through hierarchical encoding. The diffusion model construction module is used to construct a diffusion model based on the hierarchical conditional embedding, and to initially generate a car design image; The optimization module is used to perform strategy consistency evaluation based on the car design image, and to perform adaptive optimization of the mapping matrix based on the evaluation results, and output the final car design drawing.

Images