Automobile styling design method, system, electronic device and storage medium

Abstract

This application belongs to the field of product design technology and discloses a method, system, electronic device, and storage medium for automobile styling design based on a combination of visual focus sequences and artificial intelligence. The method includes: using eye tracking technology to capture the visual focus sequence of users when browsing automobile styling images to extract the key morphological features of the automobile; using a large language model to organize and summarize M types of perceptual vocabulary for automobiles, and using an entropy-BBWM-game theory hybrid analysis method to screen out N types of perceptual vocabulary with the highest weight from the M types of perceptual vocabulary; using a deep neural network prediction model based on KAT to construct a mapping relationship between automobile morphological features and N types of perceptual vocabulary, and selecting the optimal morphological feature combination based on the mapping relationship; based on the optimal morphological feature combination, generating N types of automobile styling through an SDM model. This method can solve problems such as inaccurate user emotion quantification, low reliability of morphological feature combination prediction, and insufficient efficiency of multi-source data fusion in automobile exterior design.

Description

Technical Field

[0001] The present invention relates to the technical field of product design, and in particular to an automobile styling design method, system, electronic equipment and storage medium based on combining visual focus sequence and artificial intelligence. Background Art

[0002] Current new energy vehicle form design primarily relies on Kansei Engineering techniques, translating user emotional needs into design features through emotion mapping methods (such as the Kano model, entropy-TOPSIS, and support vector regression). For example, product trends can be predicted by analyzing online user reviews, or key design elements can be identified by constructing form deconstruction tables based on visual focus sequences. In recent years, artificial intelligence technologies (such as Transformer networks) have been introduced into the design field to process morphological data and sequence dependencies to generate product form combination schemes. Existing research uses eye tracking technology to capture user visual focus sequences and combines them with machine learning models to optimize design solutions, providing a data-driven technical approach for automotive exterior design.

[0003] Traditional Kansei engineering methods rely on highly subjective weight calculations (such as AHP and KANO), resulting in low efficiency and poor generalization of emotional demand analysis. Modeling the association between visual focus sequences and morphological features is still limited by the high complexity and low interpretability of Transformer networks, prone to unstable predictions and overfitting. Furthermore, existing design processes lack a deep integration of user emotional data, visual behavior data, and morphological features, making it difficult to accurately capture dynamic aesthetic preferences, hindering the personalization and sustainable optimization of design solutions. Summary of the Invention

[0004] The purpose of the embodiments of this application is to provide an automotive styling design method, system, electronic device, and storage medium based on visual focus sequences and improved artificial intelligence models. By comprehensively calculating sentiment weights through entropy-BBWM-game theory and combining it with the Kolmogorov-Arnold network (KAN), the robustness and interpretability of the Transformer are enhanced. This allows for efficient mapping of user perceptual needs to morphological features, ultimately forming an innovative automotive design framework that balances aesthetic preferences and sustainability. This approach aims to address issues such as inaccurate user sentiment quantification, low reliability in predicting morphological feature combinations, and inefficient multi-source data fusion in automotive exterior design.

[0005] In order to solve the above technical problems, this application is implemented as follows:

[0006] In a first aspect, an embodiment of the present application provides a bionic design method for ceramic products, the method comprising:

[0007] Eye tracking technology is used to capture the user's visual focus sequence when browsing car styling images to extract the key morphological features of the car;

[0008] Using a large language model to organize and summarize M categories of emotional vocabulary related to automobiles, and using an entropy-BBWM-game theory hybrid analysis method to filter out N categories of emotional vocabulary with the highest weights from the M categories of emotional vocabulary;

[0009] Constructing a mapping relationship between the vehicle morphological features and the N types of perceptual vocabulary based on the KAT deep neural network prediction model, and selecting the optimal morphological feature combination according to the mapping relationship;

[0010] Based on the optimal combination of morphological features, N car shapes are generated through the SDM model.

[0011] As an optional implementation of the first aspect of the present application, the steps of using eye tracking technology to capture the visual focus sequence of a user when browsing car styling pictures to extract the key morphological features of the car include: tracking the eye movement trajectory of the user when observing the three views of the car based on the principle of corneal reflection and recording the gaze point transfer path to determine the visual focus sequence; according to the visual focus sequence, prioritizing the key morphological elements of the car to obtain the key morphological features of the car.

[0012] As an optional implementation of the first aspect of the present application, M types of emotional vocabulary for automobiles are sorted and summarized with the help of a large language model, and N types of emotional vocabulary with the highest weights are screened out from the M types of emotional vocabulary through an entropy-BBWM-game theory hybrid analysis method. The BBWM is a decision analysis method based on Bayesian statistics, which specifically includes: determining a set of evaluation indicators for emotional vocabulary; determining the best indicator and the worst indicator from the set of evaluation indicators based on historical experience data; constructing a first comparison matrix formed by the preference degree of the best indicator relative to all other indicators, and a second comparison matrix formed by the preference degree of all other indicators relative to the worst indicator; based on the Bayesian statistical method, the first comparison matrix and the second comparison matrix are modeled by multinomial distribution and Dirichlet distribution, and the posterior distribution is calculated using Markov chain Monte Carlo technology to determine the weight of each evaluation indicator; combined with the information entropy method, the weights of the evaluation indicators are normalized to calculate the optimal combination weight.

[0013] As an optional implementation of the first aspect of the present application, the steps of constructing a first comparison matrix formed by the preference degree of the optimal indicator relative to all other indicators, and a second comparison matrix formed by the preference degree of all other indicators relative to the worst indicator include: using numbers 1 to 9 to determine the preference degree of the optimal indicator relative to all other indicators, and determining the preference degree of all other indicators relative to the worst indicator, 1 represents equal importance, 9 represents extreme importance, and the larger the number, the greater the importance; through pairwise comparison, obtaining the first comparison matrix formed by the preference degree of the optimal indicator relative to all other indicators, and obtaining the second comparison matrix formed by the preference degree of the worst indicator relative to all other indicators.

[0014] As an optional implementation manner of the first aspect of the present application, a mapping relationship between the automobile morphological features and the N types of sensory vocabulary is constructed based on the deep neural network prediction model of KAT, and the step of selecting the optimal morphological feature combination according to the mapping relationship includes: constructing the automobile morphological features and the N types of sensory vocabulary into a data set; constructing the deep neural network prediction model of KAT based on the Kolmogorov-Arnold representation theorem as a theoretical basis; inputting the data set into the deep neural network prediction model of KAT for training to establish a mapping relationship between the automobile morphological features and the N types of sensory vocabulary; inputting the automobile morphological feature data to be tested into the trained deep neural network prediction model of KAT, and outputting a predicted sensory value; and obtaining the optimal morphological feature combination based on the predicted sensory value.

[0015] As an optional implementation of the first aspect of the present application, based on the Kolmogorov-Arnold representation theorem as a theoretical basis, the steps of constructing a deep neural network prediction model of KAT include: designing a network architecture of multiple KAN layers based on the Kolmogorov-Arnold representation theorem; stacking the multiple KAN layers together in sequence and combining them with the Transformer architecture to form a deep neural network prediction model of KAT; in the deep neural network prediction model of KAT, using the standard Transformer attention mechanism to capture global dependencies, and replacing the multi-layer perceptron MLP in the traditional Transformer with the KAN layer.

[0016] As an optional implementation of the first aspect of the present application, the steps of generating N car shapes through the SDM model based on the optimal morphological feature combination include: using NURBS surface modeling technology to construct a high-precision Class A surface model on the Rhino platform, focusing on optimizing the detailed feature structure of the car, and realizing parametric design through Grasshopper parameterization to obtain a modeling model; outputting the modeling model as a line drawing to import it into the SDM model for generation and iteration, and finally generating N car shapes; through point cloud experimental evaluation, obtaining the emotional scores of the N car shapes, and further adjusting the optimized design of the car shape according to the emotional scores.

[0017] In a second aspect, an embodiment of the present application provides an automobile styling design system, the system comprising:

[0018] A morphological feature extraction module is used to capture the user's visual focus sequence when browsing car styling images using eye tracking technology to extract the key morphological features of the car;

[0019] A perceptual vocabulary screening module is used to organize and summarize M perceptual vocabulary related to automobiles using a large language model, and to screen out N perceptual vocabulary with the highest weights from the M perceptual vocabulary using a hybrid analysis method based on entropy, BBWM, and game theory.

[0020] A feature combination selection module is used to construct a mapping relationship between the vehicle morphological features and the N types of perceptual vocabulary based on the KAT deep neural network prediction model, and select the optimal morphological feature combination according to the mapping relationship;

[0021] The automobile styling generation module is used to generate N automobile stylings through the SDM model based on the optimal morphological feature combination.

[0022] In a third aspect, an embodiment of the present application provides an electronic device comprising a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein when the program or instruction is executed by the processor, the steps of the method described in the first aspect are implemented.

[0023] In a fourth aspect, an embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the method described in the first aspect are implemented.

[0024] Compared with the prior art, the present invention proposes a method for automobile styling design, which has the following advantages:

[0025] 1. When using eye tracking technology to capture the visual focus sequence, the user's eye movement trajectory when observing the three views of the car is tracked based on the principle of corneal reflection, and the gaze point transfer path is recorded to determine the visual focus sequence; based on the visual focus sequence, the key morphological elements of the car are prioritized to obtain the key morphological features of the car.

[0026] 2. In the process of selecting perceptual vocabulary, the BBWM method includes: determining the set of evaluation indicators for perceptual vocabulary; determining the best and worst indicators based on historical experience data; constructing a preference comparison matrix; using Bayesian statistical methods and Markov Chain Monte Carlo technology to calculate the posterior distribution and determine the weight of each evaluation indicator; and combining the information entropy method to calculate the optimal combination weight.

[0027] 3. When constructing the KAT deep neural network model, we used the Kolmogorov-Arnold representation theorem as the theoretical basis, designed a network architecture with multiple KAN layers, and combined it with the Transformer architecture to form the KAT model. We used the standard Transformer attention mechanism to capture global dependencies and replaced the multi-layer perceptron (MLP) in the traditional Transformer with the KAN layer.

[0028] 4. When generating car styling, NURBS surface modeling technology is used on the Rhino platform to build a high-precision Class A surface model, and parametric design is achieved through the Grasshopper parametric plug-in. The output line drawing is imported into the SDM model for generation and iteration, ultimately generating N car styling models. The emotional scores are then determined through point cloud experimental evaluation, and the optimized design of the car styling is adjusted based on the scores. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of a method for designing an automobile shape provided by the first embodiment of the present invention;

[0030] Figure 2 3 is a schematic structural diagram of an automobile styling design system provided by the third embodiment of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0032] The terms "first," "second," and the like in the specification and claims of this application are used to distinguish similar objects and are not used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments of this application can be implemented in an order other than those illustrated or described herein. In addition, the term "and / or" in the specification and claims refers to at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0033] In order to illustrate the technical solution described in this application, specific embodiments are provided below.

[0034] Example 1

[0035] See also Figure 1 , is a flow chart of a car styling design method proposed in the first embodiment of this application, and the proposed method steps are as follows.

[0036] S1. Use eye tracking technology to capture the user's visual focus sequence when browsing car styling pictures to extract the key morphological features of the car.

[0037] In some embodiments, the eye movement trajectory of the user is tracked when observing the three views of the car based on the principle of corneal reflection and the gaze point transfer path is recorded to determine the visual focus sequence; according to the visual focus sequence, the key morphological elements of the car are prioritized to obtain the key morphological features of the car.

[0038] It's important to note that the visual focus sequence refers to the order in which subjects' eye fixations linger when observing the three views of a car. It reveals the subjects' subconscious focus when presented with complex visuals, often corresponding to key attractive features of the car's form. Eye trackers, based on the principle of corneal reflection, track user eye movements and record the path of gaze shifts to determine the visual focus sequence, thereby revealing the distribution of psychological emphasis in consumers' perception of car form. By analyzing eye movement test datasets, it is possible to accurately extract the priority ranking of key morphological elements in a car's styling, providing data support for automotive designers to build visual preference decision-making models, enabling them to systematically understand consumer aesthetic preferences and effectively match styling features with market demand through scientific design approaches.

[0039] In this step, an eye tracker based on corneal reflection tracks the user's eye movements while observing the three views of the car and records the path of gaze shifts. This allows for the precise identification of visual focus sequences, revealing the subjects' subconscious focus points when presented with complex views. These focus points often correspond to key attractive features of the car's form. Extracting visual focus sequences provides a scientific basis for analyzing the distribution of psychological emphasis in consumers' perceptions of car form. Analysis of eye movement test datasets allows for the precise prioritization of key morphological elements in car styling, helping designers systematically understand consumers' aesthetic preferences.

[0040] S2. With the help of a large language model, M categories of emotional vocabulary related to automobiles are sorted and summarized. Then, through the entropy-BBWM-game theory hybrid analysis method, N categories of emotional vocabulary with the highest weights are selected from the M categories of emotional vocabulary.

[0041] For example, before conducting computational work on perceptual vocabulary, the primary task is to extensively collect and rationally categorize perceptual vocabulary. Based on this, this study leveraged the powerful language model ChatGPT to conduct a comprehensive search and analysis of the automotive field, meticulously organizing the massive amount of perceptual vocabulary collected into M categories. To accurately select the N most highly weighted perceptual vocabulary categories, the study employed a hybrid analysis method combining BBWM (to integrate expert subjective preferences), entropy weighting (to objectively quantify data dispersion), and game theory (to dynamically balance subjective and objective weights). This approach accurately identified the N most influential perceptual vocabulary categories, providing a new methodology for sustainable automotive styling design.

[0042] Among them, it should be emphasized that BBWM is a decision analysis method based on Bayesian statistics. The specific calculation steps include: determining a set of evaluation indicators for perceptual vocabulary; determining the optimal indicator and the worst indicator from the evaluation indicator set based on historical experience data; constructing a first comparison matrix formed by the preference degree of the optimal indicator relative to all other indicators, and a second comparison matrix formed by the preference degree of all other indicators relative to the worst indicator; based on the Bayesian statistical method, the first comparison matrix and the second comparison matrix are modeled through multinomial distribution and Dirichlet distribution, and the Markov chain Monte Carlo technology is used to calculate the posterior distribution and determine the weight of each evaluation indicator; combined with the information entropy method, the weight of each evaluation indicator is normalized to calculate the optimal combination weight.

[0043] Furthermore, the numbers 1 to 9 are used to determine the preference of the optimal indicator relative to all other indicators, and the preference of all other indicators relative to the worst indicator. 1 represents equal importance, 9 represents extreme importance, and the larger the number, the greater the importance. Through pairwise comparison, the first comparison matrix formed by the preference of the optimal indicator relative to all other indicators and the second comparison matrix formed by the preference of the worst indicator relative to all other indicators are obtained.

[0044] For example, the specific steps of the decision analysis method based on Bayesian statistics include:

[0045] Step 1: Determine the evaluation index set .

[0046] Step 2: In this step, multiple evaluators determine the optimal indicator based on experience from the indicator set determined in the first step. and the worst indicator , the optimal indicator is the most important indicator or the most ideal indicator, while the worst indicator is the least important or the least ideal indicator among other criteria.

[0047] Step 3: Construct comparison criteria. Use numbers 1 to 9 to determine the preference of the optimal indicator relative to all other indicators. 1 represents equal importance, 9 represents extreme importance, and the larger the number, the greater the importance. By comparing two indicators, a comparison matrix relative to the optimal indicator is obtained. .

[0048] Step 4: Determine the preference of all other criteria relative to the worst criterion, and obtain the comparison matrix relative to the worst indicator through pairwise comparison .

[0049] Step 5: Based on the previous determination of the best and worst indicators and the construction of the comparison matrix, further use complex mathematical methods to calculate the weight of each evaluation indicator. and After being the model input and output, we need to add the multinomial distribution. The multinomial probability distribution of . Similarly, for the optimal index Modeled using multinomial distribution, but with the worst-case metric The probability distribution is opposite. In summary, the weight determination method has been transformed into probability distribution estimation, and it is necessary to establish a Bayesian hierarchical model to solve it. Suppose there are N (n=1,2…N) evaluation experts, then the Nth evaluation expert will evaluate the evaluation index according to the evaluation index. …, Determine the best and worst comparison matrix as and , the best and worst comparison matrix set of N evaluation experts is: and .set up Represents the comprehensive weight determined by all evaluation experts, and the weight set of each indicator determined by each evaluation expert Calculated, which is calculated through the joint probability distribution and According to the above formula, the probability of each random variable is calculated by the following probability rule to prove that there is clear conditional independence between the variables.

[0050] First, considering the independence of all different variables, the Bayesian method is applied to the joint probability formula, in which the distribution of each parameter is calculated based on the probability chain rule, the conditional independence of different variables, and the preference of each decision maker for the standard. and Modeling. Then, for a given , it can be expected that any are all near it. To this end, the mean and concentration parameters of the Dirichlet distribution are reparameterized. Then given of Model is the mean value of the distribution, and the closeness between the two is controlled by a non-negative index, so the above formula represents the weight vector associated with each evaluator Must be In addition, the concentration parameter needs to be modeled using a gamma distribution that satisfies the non-negativity constraint. Where a and b refer to the shape parameters of the gamma distribution. Finally, using the parameter Calculation of the uninformative Dirichlet distribution The prior distribution on the statistic is given. Given that the model does not contain a closed-form solution, the posterior distribution is calculated using Markov Chain Monte Carlo (MCMC). For MCMC sampling, the posterior is sampled and calculated using the probabilistic language Just Another Gibbs Sampler (JAGS). The model ultimately outputs the posterior distribution of the optimal weight for each evaluator and the optimal aggregate weight for all evaluators.

[0051] The tool of information entropy can be used to calculate the weight of each indicator and provide a basis for the comprehensive evaluation of multiple indicators. The specific steps are as follows.

[0052] The first step is data normalization: normalize the data for positive, negative, and appropriateness indicators to avoid dimensionality issues. The second step is calculating the weight of each indicator. The third step is calculating information entropy. The fourth step is calculating the information entropy redundancy and then calculating the indicator weights. The fifth step is constructing a combined weight vector and performing differentiation on the coefficients. Finally, the weight coefficients are obtained.

[0053] Among them, when game theory integrates subjective and objective analysis to determine weights, it mainly realizes the organic integration of subjective weights and objective weights through the steps of constructing a combined weight vector, introducing game theory ideas, and normalizing the coefficients, thereby obtaining the optimal combined weights.

[0054] In this step, large language models such as ChatGPT were first used to comprehensively search and analyze emotional vocabulary in the automotive field. This enabled the efficient and systematic analysis of 12 categories of emotional vocabulary, providing a rich and structured data foundation for subsequent analysis. Secondly, a hybrid analysis method combining entropy weighting, BBWM, and game theory was used to achieve a dynamic balance between subjective and objective weightings. The entropy weighting method objectively quantified the data dispersion, the BBWM incorporated expert subjective preferences, and the game theory further optimized the weighting distribution, thereby accurately selecting the three most influential core emotional vocabulary. This process not only improved the scientific and reliable selection of emotional vocabulary but also provided a clear emotional guide for automotive styling design, enabling designers to more accurately grasp user needs and improve the market adaptability and user satisfaction of designs. Furthermore, the BBWM, based on Bayesian statistical decision analysis, ensured the accuracy and stability of weight calculation through multinomial distribution, Dirichlet distribution, and Markov Chain Monte Carlo techniques, providing solid theoretical support for comprehensive multi-metric evaluation.

[0055] S3. Based on the KAT deep neural network prediction model, a mapping relationship between automobile morphological features and N types of perceptual vocabulary is constructed, and the optimal morphological feature combination is selected according to the mapping relationship.

[0056] In some embodiments, automobile morphological features and N types of perceptual vocabulary are constructed into a data set; based on the Kolmogorov-Arnold representation theorem as a theoretical basis, a deep neural network prediction model of KAT is constructed; the data set is input into the deep neural network prediction model of KAT for training to establish a mapping relationship between automobile morphological features and N types of perceptual vocabulary; the morphological feature data of the automobile to be tested is input into the trained deep neural network prediction model of KAT, and a predicted perceptual value is output; and an optimal morphological feature combination is obtained based on the predicted perceptual value.

[0057] Furthermore, based on the Kolmogorov-Arnold representation theorem, a network architecture with multiple KAN layers is designed. Multiple KAN layers are stacked together in sequence and combined with the Transformer architecture to form KAT's deep neural network prediction model. In KAT's deep neural network prediction model, the standard Transformer attention mechanism is used to capture global dependencies, and the multi-layer perceptron (MLP) in the traditional Transformer is replaced with KAN layers.

[0058] As described in the example of step S2, by combining the entropy weight method, BBWM, and game theory, three emotional words with the highest weights are selected from the 12 categories of emotional word representatives.

[0059] For example, we then used a morphological analysis table generated from the eye-tracking experiment to construct a dataset containing 127 vehicle models and their emotional values. We then used KAT to map the NEV-SUV model to user emotions, thereby predicting the optimal morphological feature combination corresponding to the highest emotional value.

[0060] In this step, large language models such as ChatGPT were used to comprehensively search and analyze emotional vocabulary in the automotive field. This enabled the efficient and systematic analysis of 12 categories of emotional vocabulary, providing a rich and structured data foundation for subsequent analysis. Secondly, a hybrid analysis method combining entropy weighting, BBWM, and game theory was used to achieve a dynamic balance between subjective and objective weightings. The entropy weighting method objectively quantified the data's dispersion, the BBWM incorporated expert preferences, and game theory further optimized the weighting distribution, resulting in the precise selection of the three most influential core emotional vocabulary. This process not only improved the scientific and reliable nature of the emotional vocabulary selection process but also provided a clear emotional guide for automotive styling design, enabling designers to more accurately grasp user needs and improve the market adaptability and user satisfaction of their designs. Furthermore, the BBWM, based on Bayesian statistical decision analysis, ensured the accuracy and stability of weight calculation through multinomial distribution, Dirichlet distribution, and Markov Chain Monte Carlo techniques, providing solid theoretical support for comprehensive multi-metric evaluation.

[0061] S4. Generate N car shapes through the SDM model based on the optimal combination of morphological features.

[0062] In some embodiments, NURBS surface modeling technology is used to construct a high-precision Class-A surface model on the Rhino platform, focusing on optimizing the detailed feature structure of the car, and parametric design is achieved through Grasshopper parameterization to obtain a modeling model; the modeling model is output as a line drawing to be imported into the SDM model for generation and iteration, and finally N car shapes are generated; through point cloud experimental evaluation, the emotional scores of the N car shapes are obtained, and the optimized design of the car shapes is further adjusted according to the emotional scores.

[0063] Specifically, the SDM was used to design and generate N car models, which were then evaluated and scored highly in actual point cloud experiments, allowing the abstract emotional intention to be expressed in a concrete, physical form. Once KAT predicted the optimal morphological feature combination corresponding to the highest perceptual value, the combination with the highest evaluation value for each sentiment word was found. The SDM was then used to conduct actual creation, ultimately obtaining scores in the point cloud experiment evaluation phase, achieving the precise translation of the abstract emotional intention into a concrete, physical form. The specific implementation process of the point cloud test is as follows.

[0064] First, in the PyCharm development environment, use the Open3D library to import the NEV-SUV 3D model (.obj format). First, ensure the Open3D library is installed. If not, install it using the command pip install open3d . Next, import the Open3D library and visualize it using the draw_geometries function.

[0065] Next, we need to accurately calculate the point cloud density of the entire 3D model. We can use a distance-based approach to estimate the density of the point cloud by calculating the average of the nearest neighbor distances of each point in the point cloud. Specifically, we use Open3D to obtain the nearest neighbor distance for each point, and then calculate the average of these distances to obtain the point cloud density.

[0066] Finally, within the 3D model, a specific screening algorithm was used to select cross sections. Data filtering techniques were then used to discard all other surfaces, simplifying the complex 3D model into a 2D plane for easier analysis. To obtain the most representative data, an area maximization algorithm was further employed to select the cross sections with the largest area, allowing for accurate calculation of the final height. This process was then applied to batches of NEV-SUV models with high market sales to verify that the resulting model heights were within a reasonable range.

[0067] In this step, based on the optimal combination of morphological features, N car shapes were generated through the SDM model, and a high-precision Class A surface model was constructed using NURBS surface modeling technology and Grasshopper parametric design. This has significant technical effects: First, the SDM model combined with parametric design enables rapid generation and iteration of car shapes, significantly improving design efficiency while ensuring the diversity and innovation of the shapes. Secondly, the N generated car shapes were evaluated for emotional scores through point cloud experimental evaluation, transforming abstract user emotional intentions into concrete physical expressions, achieving a precise match between emotion and design, and improving the market adaptability and user satisfaction of the design. In addition, with the help of the Open3D library, the 3D model was subjected to point cloud density calculation and cross-section screening, further optimizing the model details and ensuring the scientificity and rationality of the design results.

[0068] Example 2

[0069] The following takes the new energy SUV (NEV-SUV) car styling design as an example, and combines the accompanying drawings and product renderings to further illustrate the design process of the method of the present invention, thereby verifying the effectiveness of the method proposed in the present invention. The steps of the proposed method are as follows.

[0070] First, the research team delved into mainstream automotive information websites, such as "Dongchedi" and "Autohome," to extensively collect full-view information from the front, rear, and specific angles of view of new energy SUVs on sale. Due to the wide range of data sources and different imaging conditions, professional software was used to process the collected images to remove background interference, unify the image size, and correct for distortion caused by the shooting angle. The research team used the Remove website to batch process three views of 127 cars to ensure that the images were background-free and of uniform size. This meticulous pre-processing provided high-quality and consistent image data for subsequent eye movement experiments. Through this process, a dataset covering full-view images from the front, rear, and specific angles of view of new energy SUVs was successfully constructed, laying the foundation for subsequent research on accurate vehicle morphology ranking based on eye tracking.

[0071] Next, an eye-tracking experiment was designed to assess users' visual perception. The experiment first screened six images from 127 samples that significantly affected users' emotions and standardized their resolution and dimensions. Ten participants with NEV-SUV driving experience and five industrial design researchers were recruited to ensure that no text prompts interfered. Eye movement data was collected using a TobiiPro Glass 3 eye tracker and a 21-inch monitor. Tobii Pro software analyzed and generated heat maps and eye movement trajectories. During the experiment, participants stared at the center of the screen. One image was shown every 15 seconds, for a total of six images, each shown twice. Finally, a superimposed heat map and eye movement trajectory were derived. The experimental results showed that participants primarily focused on the front and wheel hub of the NEV-SUV.

[0072] Before building the NEV-SUV sentiment mapping model, the research team used ChatGPT to systematically collect sentimental vocabulary. Through a multi-dimensional prompt word design, they collected a large number of positive sentiment words describing the NEV-SUV's appearance and driving experience. These words were then categorized into 12 categories: "dynamic," "stylish," "elegant," "luxurious," "refined," "unique," "classic," "cool," "simple," "understated," "technological," and "powerful." This process ensured comprehensive coverage of sentimental dimensions and provided a rich vocabulary foundation for subsequent sentiment analysis.

[0073] The research team then selected seven experts and subjectively assigned weights to each indicator using the Bayesian Optimal Weighting Method (BBWM). Based on their judgment, the experts selected the best and worst indicators, respectively. Based on these preference relationships, they constructed a BO (Best-to-Others) matrix and an OW (Others-to-Worst) matrix. This method effectively addressed the inconsistencies in expert opinions and ensured the scientific and objective nature of the weighting results.

[0074] Based on this, the research team designed a questionnaire, incorporating the collected NEV-SUV samples and sentimental vocabulary, and invited 120 participants with design backgrounds to rate them. Using a seven-point trait scale, they combined the collected means with the morphological structure table to construct a sentiment evaluation matrix, providing a foundation for subsequent analysis of the relationship between morphological characteristics and sentiment.

[0075] To establish a mapping relationship between user emotions and NEV-SUV morphology, the research team constructed a deep neural network prediction model based on KAT. First, the morphological features of the NEV-SUV were normalized to ensure consistent dimensionality of the feature data and avoid training bias. Next, a KAT model based on the improved Transformer architecture was constructed, replacing the MLP layer with the KAN layer and retaining the multi-head attention mechanism to accommodate complex emotional and morphological data analysis. During training, the dataset was divided into a training set (70%) and a test set (30%). Using the Adam optimizer and the mean squared error loss function, the model was iteratively trained until the accuracy was met, successfully establishing a mapping relationship between user emotions and NEV-SUV morphology.

[0076] To further analyze the relationship between morphological features and user emotions, the research team used an exhaustive method to calculate all possible morphological feature combinations. Ultimately, they identified 117,649 different morphological feature combinations and predicted them for the three emotional dimensions of "fashion," "dynamic," and "luxury." This analysis revealed that specific morphological feature combinations maximized user experiences of "fashion," "dynamic," and "luxury," providing data support for design decisions.

[0077] Finally, based on the KE theoretical framework, the research team implemented innovative styling design practices in the new energy SUV sector. Using the entropy-BBWM-game theory approach, they selected three core perceptual concepts: "fashion," "dynamic," and "luxury" as design guidelines and established a corresponding emotional evaluation system. During the digital modeling phase, they used NURBS surface modeling technology to construct a high-precision Class-A surface model on the Rhino 7.0 platform. They focused on optimizing the light guide structure of the through-type taillights and implemented a gradient algorithm for the grille unit using the Grasshopper parametric plug-in. During the rendering phase, the model line drawings were imported into the SDM model for generation and iterative optimization, ultimately completing the innovative styling design of the new energy SUV.

[0078] Example 3

[0079] See also Figure 2 , which is a schematic diagram of the structure of an automobile styling design system proposed in the third embodiment of the present application, the system includes:

[0080] The morphological feature extraction module 100 is used to capture the visual focus sequence of the user when browsing the car modeling picture using eye tracking technology to extract the key morphological features of the car;

[0081] The perceptual vocabulary screening module 200 is used to organize and summarize M perceptual vocabulary related to automobiles using a large language model, and screen out N perceptual vocabulary with the highest weights from the M perceptual vocabulary using a hybrid analysis method based on entropy, BBWM, and game theory.

[0082] A feature combination selection module 300 is configured to construct a mapping relationship between the vehicle morphological features and the N types of perceptual vocabulary based on the KAT deep neural network prediction model, and select an optimal morphological feature combination based on the mapping relationship;

[0083] The car shape generation module 400 is used to generate N car shapes through the SDM model based on the optimal morphological feature combination.

[0084] In the embodiments of the present application, an automotive styling design system can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, a mobile electronic device can be a mobile phone, tablet computer, laptop computer, PDA, in-vehicle electronic device, wearable device, ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), while a non-mobile electronic device can be a server, network attached storage (NAS), personal computer (PC), etc., without specific limitations in the embodiments of the present application.

[0085] In the embodiment of the present application, a car styling design system can be a device having an operating system. The operating system can be an Android operating system, an iOS operating system, or other possible operating systems, which are not specifically limited in the embodiment of the present application.

[0086] The automobile styling design system provided in the embodiment of the present application can realize Figure 1 To avoid repetition, each process of implementing a method for automobile styling design in the method embodiment will not be described here.

[0087] Optionally, an embodiment of the present application also provides an electronic device, including a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, each process of the above-mentioned embodiment of the automobile styling design method is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be described here.

[0088] An embodiment of the present application also provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the various processes of the embodiment of the above-mentioned automobile styling design method are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0089] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer-readable storage medium, such as a computer read-only memory (ROM), random access memory (RAM), a magnetic disk, or an optical disk.

[0090] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element. In addition, it should be noted that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in the opposite order according to the functions involved. For example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0091] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a more preferred embodiment. Based on this understanding, the technical solution of this application, or the part that contributes to the existing technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of this application.

[0092] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of this application, ordinary technicians in this field can also make many forms without departing from the purpose of this application and the scope of protection of the claims, all of which are within the protection of this application.

Claims

1. A method for designing automobile styling, characterized in that: include: Eye tracking technology is used to capture the user's visual focus sequence when browsing car styling images to extract the key morphological features of the car; M categories of car-related sentimental vocabulary are summarized and sorted out with the help of a large language model. Then, N categories of sentimental vocabulary with the highest weights are selected from the M categories of sentimental vocabulary through a hybrid analysis method based on entropy, BBWM, and game theory. The BBWM is a decision analysis method based on Bayesian statistics, which specifically includes: determining a set of evaluation indicators for sentimental vocabulary; determining the optimal indicator and the worst indicator from the set of evaluation indicators based on historical experience data; constructing a first comparison matrix formed by the degree of preference of the optimal indicator relative to all other indicators, and a second comparison matrix formed by the degree of preference of all other indicators relative to the worst indicator; based on the Bayesian statistical method, the first and second comparison matrices are modeled using multinomial distribution and Dirichlet distribution, and the posterior distribution is calculated using Markov chain Monte Carlo technology to determine the weight of each evaluation indicator; and combining the information entropy method to normalize the weights of the evaluation indicators and calculate the optimal combination weight. A mapping relationship between the vehicle morphological features and the N types of perceptual vocabulary is constructed based on a KAT deep neural network prediction model, and an optimal morphological feature combination is selected based on the mapping relationship, including: constructing a data set with the vehicle morphological features and the N types of perceptual vocabulary; constructing a KAT deep neural network prediction model based on the Kolmogorov-Arnold representation theorem; inputting the data set into the KAT deep neural network prediction model for training to establish a mapping relationship between the vehicle morphological features and the N types of perceptual vocabulary; inputting the morphological feature data of the vehicle to be tested into the trained KAT deep neural network prediction model to output a predicted perceptual value; and obtaining an optimal morphological feature combination based on the predicted perceptual value; Based on the optimal combination of morphological features, N car shapes are generated through the SDM model.

2. The automobile styling design method according to claim 1, characterized in that: The steps for using eye tracking technology to capture the user's visual focus sequence when browsing car styling images and extracting the car's key morphological features include: Based on the principle of corneal reflection, the user's eye movement trajectory when observing the three views of the car is tracked and the gaze point transfer path is recorded to determine the visual focus sequence; The key morphological elements of the automobile are prioritized according to the visual focus sequence to obtain key morphological features of the automobile.

3. The automobile styling design method according to claim 1, characterized in that: The steps of constructing a first comparison matrix formed by the preference degree of the optimal indicator relative to all other indicators, and a second comparison matrix formed by the preference degree of all other indicators relative to the worst indicator include: Use numbers from 1 to 9 to determine the preference of the best indicator relative to all other indicators, and to determine the preference of all other indicators relative to the worst indicator, with 1 representing equal importance and 9 representing extreme importance. The larger the number, the greater the importance. By comparing two indicators in pairs, a first comparison matrix formed by the preference degree of the optimal indicator relative to all other indicators is obtained, and a second comparison matrix formed by the preference degree of the worst indicator relative to all other indicators is obtained.

4. The automobile styling design method according to claim 1, characterized in that: Based on the Kolmogorov-Arnold representation theorem, the steps to construct KAT's deep neural network prediction model include: According to the Kolmogorov-Arnold representation theorem, a network architecture with multiple KAN layers is designed; Multiple KAN layers are stacked together in sequence and combined with the Transformer architecture to form the KAT deep neural network prediction model; In the KAT's deep neural network prediction model, the standard Transformer attention mechanism is used to capture global dependencies, and the multi-layer perceptron (MLP) in the traditional Transformer is replaced by the KAN layer.

5. The automobile styling design method according to claim 1, characterized in that: Based on the optimal combination of morphological features, the steps of generating N car models through the SDM model include: Using NURBS surface modeling technology on the Rhino platform to build a high-precision Class A surface model, focusing on optimizing the detailed feature structure of the car, and implementing parametric design through Grasshopper parameterization to obtain the model; Outputting the modeling model into a line drawing to be imported into the SDM model for generation and iteration, and finally generating N car models; Through point cloud experimental evaluation, emotional scores of the N car models are obtained, and the optimal design of the car models is further adjusted according to the emotional scores.

6. An automobile styling design system, characterized in that: The system comprises: A morphological feature extraction module is used to capture the user's visual focus sequence when browsing car styling images using eye tracking technology to extract the key morphological features of the car; A perceptual vocabulary screening module is used to organize and summarize M types of perceptual vocabulary for automobiles with the help of a large language model, and screen out N types of perceptual vocabulary with the highest weights from the M types of perceptual vocabulary through an entropy-BBWM-game theory hybrid analysis method. The BBWM is a decision analysis method based on Bayesian statistics, which specifically includes: determining a set of evaluation indicators for perceptual vocabulary; determining the optimal indicator and the worst indicator from the evaluation indicator set based on historical experience data; constructing a first comparison matrix formed by the preference degree of the optimal indicator relative to all other indicators, and a second comparison matrix formed by the preference degree of all other indicators relative to the worst indicator; based on the Bayesian statistical method, the first comparison matrix and the second comparison matrix are modeled using multinomial distribution and Dirichlet distribution, and the posterior distribution is calculated using Markov chain Monte Carlo technology to determine the weight of each evaluation indicator; and in combination with the information entropy method, the weights of the evaluation indicators are normalized to calculate the optimal combination weight. A feature combination selection module is used to construct a mapping relationship between the vehicle morphological features and the N types of perceptual vocabulary based on the KAT deep neural network prediction model, and select an optimal morphological feature combination based on the mapping relationship, including: constructing the vehicle morphological features and the N types of perceptual vocabulary into a data set; constructing the KAT deep neural network prediction model based on the Kolmogorov-Arnold representation theorem as a theoretical basis; inputting the data set into the KAT deep neural network prediction model for training to establish a mapping relationship between the vehicle morphological features and the N types of perceptual vocabulary; inputting the morphological feature data of the vehicle to be tested into the trained KAT deep neural network prediction model, and outputting a predicted perceptual value; and deriving an optimal morphological feature combination based on the predicted perceptual value; The automobile styling generation module is used to generate N automobile stylings through the SDM model based on the optimal morphological feature combination.

7. An electronic device, characterized in that: The method comprises a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of the automobile styling design method according to any one of claims 1 to 5 are implemented.

8. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the automobile styling design method according to any one of claims 1 to 5 are implemented.

Images