Automobile modeling intelligent screening method based on measurement aesthetic entropy evaluation model

Abstract

The invention discloses an intelligent automobile modeling screening method based on a measurement aesthetic entropy evaluation model, and relates to the technical field of automobile design. The method comprises the steps of collecting automobile modeling image data, performing preprocessing operations such as gray processing, background cutting and standardization processing on an image, and constructing a line draft image set for automobile modeling beauty value calculation; calculating an automobile modeling beauty degree value based on a measurement aesthetic entropy evaluation model; a pre-trained ResNet50 convolutional neural network model is used for training, a mean square error (MSE) is used as a loss function, and an Adam optimizer is used for parameter updating until the loss function converges; and finally, intelligently screening the automobile modeling schemes according to the trained model. According to the method, objective and quantitative evaluation of the automobile modeling scheme is realized, the efficiency and accuracy of automobile design are improved, a new design tool and method are provided for automobile designers, and innovation and development of automobile intelligent design are promoted.

Description

Technical Field

[0001] The invention relates to the technical field of automobile design, and in particular to an intelligent screening method for automobile shapes based on a quantitative aesthetic entropy evaluation model. Background Art

[0002] With the rapid development of the automobile industry, automobile design has gradually shifted from traditional computer-aided design to intelligent design. The traditional automobile design process usually relies on the designer's experience and intuition, lacks scientific and quantitative evaluation criteria, and leads to low design efficiency and strong subjectivity. In recent years, the rapid development of artificial intelligence technology has brought new opportunities for automobile design, especially the outstanding performance of convolutional neural network (CNN) in image recognition and classification, making it an important tool for quantitative aesthetic convolutional neural network.

[0003] However, most of the existing intelligent aesthetic evaluation methods for automobile styling are based on simple visual feature extraction and classification, lacking an objective measurement of the beauty of automobile styling. Aesthetic metric theory and information entropy theory provide a theoretical basis for quantitative evaluation of automobile styling. Aesthetic metric theory describes beauty through mathematical models, while information entropy theory is used to evaluate the orderliness and complexity of information. Combining these two theories can more comprehensively evaluate the aesthetic value of automobile styling. Summary of the invention

[0004] The purpose of this invention is to propose an intelligent screening method for automobile styling based on a quantitative aesthetic entropy evaluation model. By combining aesthetic measurement theory, information entropy theory and complexity theory, a quantitative aesthetic entropy evaluation model is constructed to achieve scientific quantification and objective evaluation of the beauty of automobile styling.

[0005] To achieve the above purpose, the present invention proposes an intelligent screening method for automobile styling based on a quantitative aesthetic entropy evaluation model, and the specific steps are as follows:

[0006] S1. Car styling image data collection and preprocessing, constructing a car styling image dataset for measuring aesthetic convolutional neural networks and a line drawing atlas for calculating car styling beauty values;

[0007] S2. Based on the Birkhoff aesthetic measurement theory model, entropy evaluation theory model and complexity theory model, a quantitative aesthetic entropy evaluation model is constructed to calculate the beauty value of automobile styling;

[0008] S3. Constructing a quantitative aesthetic convolutional neural network model based on quantitative aesthetic entropy evaluation

[0009] S4. Validation of the convolutional neural network model for quantitative aesthetics.

[0010] Preferably, in S1, the preprocessing operation includes grayscale processing, background clipping batch processing, standardization processing, automobile modeling structure line extraction processing and noise reduction processing.

[0011] Preferably, in S2, the specific calculation steps of the automobile styling beauty value are as follows:

[0012] S21. Measure the entropy redundancy of the three-dimensional modeling lines of the automobile based on the entropy evaluation model and calculate the aesthetic order of automobile modeling;

[0013] S22. Calculate the complexity of automobile styling aesthetics based on the complexity theory model;

[0014] S23. Based on the quantitative aesthetic entropy evaluation model, the automobile styling aesthetic order and the automobile styling aesthetic complexity are comprehensively considered to calculate the automobile styling beauty value.

[0015] Preferably, in S21, the calculation formula of the automobile styling aesthetic order is as follows:

[0016]

[0017] Among them, O is the aesthetic order of automobile styling, O h For each view’s morphological order, W h is the weight value for each view, h is the four views of the car shape;

[0018] The morphological order of each view is O h , and its calculation formula is as follows:

[0019]

[0020] Among them, S j is the information entropy of structure line j, W j is the weight value of the modeling structure line, f j (x i ) is the car modeling curve function, P ji is the ratio of the derivative value of point i on the jth structure line to the sum of the derivative values ​​of all points of the mth differential on the structure line, k = (ln m) -1 , which can satisfy 0≤S j ≤1; m is the number of points on the curve with different absolute values ​​of derivatives, i is an integer, j is an integer, and n is an integer.

[0021] Preferably, the weight value W of the modeling structure line j The Likert nine-level scale is used to weight the car styling structure line, and the calculation formula is as follows:

[0022]

[0023] Among them, P jis the average score of each modeling structure line, and n is the total number of modeling structure lines in each view.

[0024] Preferably, in S22, the formula of the complexity theory model is as follows:

[0025] C = r × E × t;

[0026] Among them, C is the aesthetic complexity of automobile styling, r is the simplest order of automobile styling, E is the dissipation of automobile styling information energy, and t is the slow growth rate.

[0027] Preferably, in S23, the quantitative aesthetic entropy evaluation model formula is as follows:

[0028]

[0029] Among them, M is the beauty value of the car's styling.

[0030] Preferably, in S3, the quantitative aesthetic convolutional neural network model uses a pre-trained ResNet50 network as a basic model and uses mean square error MSE as a loss function.

[0031] Preferably, the calculation formula of the loss function is as follows:

[0032]

[0033] Where n is the total number of samples, y λ 、M λ is the true label data of the beauty value of the λth sample, The beauty value prediction data for the λth sample.

[0034] Preferably, the quantitative aesthetic convolutional neural network model uses an Adam optimizer to update model parameters, and the model training parameters include the number of iterations epoch, the batch size batch_size and the learning rate.

[0035] Therefore, the present invention proposes an intelligent screening method for automobile styling based on a quantitative aesthetic entropy evaluation model, and its beneficial effects are as follows:

[0036] (1) The present invention combines aesthetic measurement theory, information entropy theory and complexity theory to construct a quantitative aesthetic entropy evaluation model, realizes scientific quantification and objective evaluation of the beauty of automobile styling, and improves the scientificity and reliability of automobile design.

[0037] (2) The present invention uses a convolutional neural network (CNN) to intelligently screen automobile styling schemes, which greatly improves the screening efficiency of automobile styling schemes, accelerates the design cycle, and reduces design costs.

[0038] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of an intelligent screening method for automobile styling based on a quantitative aesthetic entropy evaluation model of the present invention;

[0040] Figure 2 It is a schematic diagram of the effect of the line draft processing process of an intelligent screening method for automobile styling based on a quantitative aesthetic entropy evaluation model of the present invention;

[0041] Figure 3 The corresponding position distribution diagram of automobile styling evaluation indexes of an automobile styling intelligent screening method based on a quantitative aesthetic entropy evaluation model of the present invention;

[0042] Figure 4 The network structure of ResNet50 is a method for intelligently screening automobile shapes based on a quantitative aesthetic entropy evaluation model of the present invention;

[0043] Figure 5 This is a method for intelligently screening automobile styling based on a quantitative aesthetic entropy evaluation model and a questionnaire for evaluating the beauty of automobile styling schemes;

[0044] Figure 6 It is a schematic diagram of data set preprocessing results in an embodiment of an intelligent screening method for automobile styling based on a quantitative aesthetic entropy evaluation model of the present invention;

[0045] Figure 7 It is a schematic diagram of the preprocessing results of the line drawing atlas calculated by the Mei degree in an embodiment of an intelligent screening method for automobile styling based on a quantitative aesthetic entropy evaluation model of the present invention;

[0046] Figure 8 It is a pseudo code diagram of ResNet50 model training in an intelligent screening method for automobile styling based on a quantitative aesthetic entropy evaluation model of the present invention;

[0047] Fig. 9 It is a pseudo code schematic diagram of a quantitative aesthetic convolutional neural network model in an automobile styling intelligent screening method based on a quantitative aesthetic entropy evaluation model of the present invention;

[0048] Fig.10 It is a line graph of the convergence of the loss function in the intelligent screening method for automobile styling based on the quantitative aesthetic entropy evaluation model of the present invention;

[0049] Fig.11 This is a schematic diagram of preprocessing pictures of a verification data set in an embodiment of an intelligent screening method for automobile styling based on a quantitative aesthetic entropy evaluation model of the present invention. DETAILED DESCRIPTION

[0050] In order to make the technical solutions, advantages and purposes of the present invention clearer, the technical solutions of the embodiments of the present invention are clearly and completely described below. The described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the described embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of this application.

[0051] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0052] like Figure 1 As shown, the present invention provides an intelligent screening method for automobile styling based on a quantitative aesthetic entropy evaluation model, and the specific steps are as follows:

[0053] S1. Car styling image data collection and preprocessing

[0054] To collect samples of automobile styling, a crawler program was written based on Python language to obtain automobile styling samples from the Dongchedi (www.dongchedi.com) online platform, and styling image data (front view, rear view, side view, top view) of different brands of sedans, SUVs and sports cars produced after 2003 were selected.

[0055] Sample set preprocessing, the present invention calculates the beauty of automobile styling by measuring the aesthetic entropy evaluation model. The content does not involve color beauty, so the sample set is grayscale preprocessed and a standardized view is constructed to eliminate the interference of redundant factors and construct a standard data set for convolutional neural network training and optimization. The standard data set after grayscale preprocessing is further processed to construct a line drawing atlas for calculating the beauty value of automobile styling. The specific process is as follows:

[0056] First, grayscale preprocessing was performed. The sample set images were imported into the cutout website (ydtai.com / remove-background) in units of car models for batch background cropping. The cropped images were imported into Photoshop software for grayscale batch processing, and the bit depth value was set to 8. The image background was uniformly standardized and grayed (R:138, G:138, B:138), and the license plate areas in the front and rear views were uniformly blocked with gray rectangles (R:202, G:202, B:202).

[0057] Then, a standardized view was constructed, and the front, rear, side and top views of the car were placed in a 1181px×1378px, 100dpi screen. The front and rear views were at the top of the screen, the side and top views had the front of the car facing left, the side view was in the middle, and the top view was at the bottom. The image sizes of the four views of the car's shape were based on the height of the car, and the image margins were uniformly set to: 39px for the top, bottom, left and right margins, and 39px for the spacing between each view.

[0058] Finally, the standard data set after grayscale preprocessing is further processed to construct a line drawing atlas for calculating the beauty value of automobile styling. The Sobel operator program (an image processing tool for edge detection that can identify boundaries and contours in automobile styling images) is implemented through MATLAB. The styling structure lines of the front, rear, side, and top views of the automobile styling are extracted in batches. The line drawing atlas of the four views is imported into Photoshop software, and the structural curves of the automobile styling are smoothed and connected. According to the list of automobile styling structure lines proposed by Li Yanlong et al., as shown in Table 1, the processed line drawings are denoised; the effect of the line drawing processing process (taking NIO ET5T as an example) is shown in the figure below. Figure 2 As shown, the car shape structure lines correspond to the corresponding views, such as Figure 3 The processed line drawing atlas is used to calculate the beauty value of the car styling, and the beauty value is used as the label of the convolutional neural network dataset.

[0059] Table 1 List of automobile styling structural lines

[0060]

[0061]

[0062] S2. Based on the Birkhoff aesthetic measurement theory model, entropy evaluation theory model and complexity theory model, a quantitative aesthetic entropy evaluation model is constructed to calculate the beauty value of automobile styling.

[0063] First, the redundancy of the entropy of the car styling structure line is calculated. The car styling line drawing is imported into the Rhino8 software. The curve tool is used to construct the car styling structure line in proportion (because the car styling is highly symmetrical, only half of the styling structure line is selected in the front view, rear view, and top view of the car styling according to the central symmetry axis), and the derivative calculation method based on the curve parameterization representation (Grasshopper plug-in) is used to calculate the derivative of the j-th structure line. The specific method steps are: select and import the j-th structure line into the Curve component; use the Divide Distance component to differentiate the structure line according to the limit equidistant to obtain discrete points; use the Derivatives component to calculate the derivatives of each point on the structure line; use the DeconstructVector component to split the derivative vector into x and y components (two-dimensional curves), calculate the derivatives of each node, use the Grasshopper plug-in to calculate the derivatives of each discrete point on the car styling structure line, and import the derivative data into the Excel software. The redundancy of the information entropy of the j-th structure line of the car styling is calculated according to the formula through the data analysis tool in the software. The formula of the data analysis tool is:

[0064]

[0065] Secondly, calculate the weight value W of the modeling structure line j , seeking the aesthetic order of automobile styling.

[0066] As shown in Table 2, the Likert nine-level scale is combined with expert evaluation (the expert team is mainly composed of professors and associate professors from the School of Design and Art, including 5 automotive design experts, 8 industrial design experts, 4 ergonomics experts, 3 aesthetic theory experts, and 5 vehicle engineering experts) to carry out the weighting of automotive styling structural lines. First, collect expert evaluation data; then calculate the average score of each styling structural line; finally, normalize the importance score of each styling structural line, and the normalization formula is:

[0067]

[0068] Among them, W j is the weight value of the modeling structure line for each view, P j is the average score of each modeling structure line, and n is the total number of modeling structure lines in each view.

[0069] Table 2 Indicator importance rating Likert nine-level scale

[0070] Points Importance 1 Not important at all 2 Very unimportant 3 Not important 4 Somewhat unimportant 5 neutral 6 Somewhat important 7 More important 8 Very important 9 Extremely important

[0071] Combined formula:

[0072]

[0073] The Excel software is used to perform weighted summation of the information entropy redundancy of each structural line and calculate the aesthetic order of each view of the automobile styling.

[0074] Then the aesthetic complexity of the car shape is calculated, and the calculation formula is as follows:

[0075] C = r × E × t;

[0076] r is the dimension of the vector space of the vehicle's structural line. Based on the principle of the simplest order of complexity, the front view, rear view, side view, and top view are Figure 4 The calculation of the morphological order of the automobile appearance is carried out using each view, so r=3; since the present invention only involves the beauty measurement of the automobile appearance, and does not involve the iteration and improvement of the automobile appearance, E=1, t=1.

[0077] Finally, the automobile styling beauty value is calculated according to the quantitative aesthetic entropy evaluation model, and the formula is as follows:

[0078]

[0079] Among them, W h The importance weight of each view is assigned using the Likert scale combined with expert evaluation.

[0080] S3. Construct a quantitative aesthetic convolutional neural network model based on quantitative aesthetic entropy evaluation and train it.

[0081] 1. Data standardization. Before training begins, standardizing the input data is a necessary step to ensure that the model can converge effectively. For the car beauty evaluation task, in order to avoid the scale difference of the input value affecting the training effect of the convolutional neural network, the following normalization formula is used:

[0082]

[0083] This method aims to ensure balanced input of data and prevent small numerical differences from adversely affecting model performance. By scaling each input value to the square of its maximum value, we can reduce the impact of data scale on model training stability.

[0084] 2. Initial settings of the model. Figure 4 As shown, the present invention uses the pre-trained ResNet50 network as the basic model. During the model training process, the mean square error (MSE) is used as the loss function to measure the difference between the predicted value and the true value, where the predicted value is the output of the model and the true value is the calculated value of the quantitative aesthetic entropy evaluation model; the Adam optimizer is selected to update the model parameters, the epoch is set to 300, the batch_size is set to 32, the learning rate is set to 0.0001, and the detailed configuration of the model hyperparameters is shown in Table 3.

[0085] Table 3 Model hyperparameter configuration table

[0086]

[0087] 3. Model training, evaluation and update. Input data to the model for forward propagation and back propagation. Forward propagation calculates the output prediction value of the model. The back propagation algorithm calculates the gradient of the model parameters through the loss function until the loss function converges. The optimizer adjusts and updates the model parameters according to the direction and size of the gradient.

[0088] 4. Model preservation and deployment: When the training process reaches the expected convergence effect, save the trained model weights and structure for the subsequent automated calculation and application of beauty evaluation tasks.

[0089] S4. Validation of the convolutional neural network model for quantitative aesthetics.

[0090] Taking the undergraduate class assignment of automobile styling scheme design taught by me as the verification data, 9 automobile styling hand-drawn design schemes were selected as the basic samples through high-quality screening, and the intelligent design scheme of automobile styling was generated with the help of AIGC technology to construct the verification data set of the quantitative aesthetic entropy evaluation model. Specific process: Apply the Stable Diffusion algorithm, import the automobile styling Lora model selected from the liblib (www.liblib.art) website into the Stable Diffusion model library, select the Chilloutmix-Ni-pruned-fp16-fix model in combination with its webui interface, adjust the prompt words and related parameters, randomly generate preliminary car styling schemes, run controlnet to upload the automobile styling hand-drawn design scheme view, and control the morphology to further generate relatively optimized schemes, select the ideal scheme, and automatically iterate and refine the 4x-ultrasharp algorithm under the high-resolution repair plug-in to complete the scheme design.

[0091] The validation data set is used to validate the econometric aesthetic convolutional neural network model, and the validation data is collated. Specific process: Run the intelligent screening program for automobile styling schemes through PyTorch, save the validation data set to the folder of images to be evaluated, use the cd command in the terminal window to switch to the directory where the econometric aesthetic convolutional neural network model is located and run it, enter the pythontest.py command in the terminal window to run the program and output the results.

[0092] The validation data set was used to conduct a questionnaire survey on the aesthetic evaluation of automobile styling solutions. The specific process is as follows:

[0093] The first step is questionnaire design; Figure 5As shown, the information designed for the questionnaire mainly includes: the subjects' gender, occupation (including: students, teachers), car styling scheme number and picture, and beauty level (Likert nine-level scale beauty level: extremely unattractive, very unattractive, unattractive, somewhat unattractive, neutral, somewhat beautiful, beautiful, very beautiful, and extremely beautiful).

[0094] The second step is to identify the subjects; 150 teachers and students from the School of Design and Art were selected as the subjects.

[0095] The third step is to distribute the automobile styling scheme aesthetic evaluation questionnaire to the subjects as a reward feedback, and ask each subject to evaluate the aesthetics of each styling scheme based on his or her true feelings.

[0096] The fourth step is to screen out invalid questionnaires (including missing data, uniform answers, and failure to fill in as required), organize the survey data, and use Cronbach's Alpha coefficient to analyze the data reliability. 0 means unreliable; 1 means completely reliable; less than 0.6 means poor reliability; greater than 0.7 means good reliability. The classic Cronbach's Alpha formula is:

[0097]

[0098] Among them, α represents Cronbach's Alpha value, N represents the number of solutions in the questionnaire, is the variance of the ith solution in the questionnaire, is the variance of the total score of all options for each respondent.

[0099] The fifth step is to calculate the aesthetic evaluation score of each car styling design scheme and sort them from high to low based on the score.

[0100] Finally, the verification results of the quantitative aesthetic convolutional neural network model are compared with the results of the automobile styling scheme beauty evaluation questionnaire survey to verify the accuracy of the intelligent screening of the quantitative aesthetic convolutional neural network model. Specific method: Use the Spearman rank correlation coefficient to test the verification results (-1 indicates a complete negative correlation, 0 indicates a complete irrelevance, and 1 indicates a complete positive correlation). The formula of the Spearman rank correlation coefficient is:

[0101]

[0102] Where: s is the Spearman rank correlation coefficient, d i is the rank difference of each pair of observations (i.e., for each i, the difference in its rank in the two variables is calculated), and n is the total number of observations.

[0103] Example 1

[0104] 1. Automobile styling sample collection and preprocessing

[0105] According to the car styling sample collection and preprocessing method in S1, a total of 10,504 styling pictures of 2,626 cars were obtained. All car models were randomly divided into training set, test set and validation set in a ratio of 8:1:1: the training set includes 2,100 models, the test set includes 263 models, and the validation set includes 263 models. There is no overlap between the divided data sets. The car styling pictures in the data set are integrated by car model, and grayscale preprocessing and standardization are performed to construct a standard data set for convolutional neural network training and optimization (2,626 pictures, such as Figure 6 As shown); further process the standard data set after grayscale preprocessing to construct a line drawing data set for calculating the beauty value of automobile styling (2626 groups, such as Figure 7 shown).

[0106] 2. Calculation of the beauty of car styling

[0107] The weight values ​​of the car styling structure lines of each view are calculated in combination with the view weights, as shown in Table 4; the importance weight values ​​of the four views of car styling are shown in Table 5; and finally, the car styling beauty values ​​of 2626 models are calculated, as shown in Table 6.

[0108] Table 4 Importance weight of automobile styling structure lines

[0109]

[0110]

[0111]

[0112] Table 5 Importance weights of the four views of automobile styling

[0113] view Weight Side View 0.41 Top view 0.22 Front View 0.19 Rear View 0.18

[0114] Table 6 Calculation results of automobile styling beauty value

[0115]

[0116]

[0117] 3. Construction and training of the convolutional neural network model for quantitative aesthetics

[0118] like Figure 8-9 As shown in Figure 2, the pre-trained ResNet50 network is used as the basic model to build a quantitative aesthetic convolutional neural network model. Fig.10 As shown in the figure, during the model training process, the mean square error (MSE) is used as the loss function, and the model is trained until the loss function converges.

[0119] 4. Verification and analysis of the training results of the quantitative aesthetic convolutional neural network model

[0120] First, a validation dataset for the econometric aesthetic convolutional neural network model was constructed and preprocessed in combination with the S1 method design, such as Fig.11 As shown in Table 7. Secondly, the verification data set is used to verify the quantitative aesthetic convolutional neural network model, and the result data is output, as shown in Table 7. Thirdly, the verification data set is used to conduct a questionnaire survey on the beauty evaluation of automobile styling schemes, and reliability analysis is carried out to organize relevant data, as shown in Table 8. Finally, the verification results of the quantitative aesthetic convolutional neural network model are compared with the results of the questionnaire survey on the beauty evaluation of automobile styling schemes. The analysis results are shown in Table 9. The Spearman rank correlation coefficient of the two is 0.9667, which proves that the results of the intelligent screening of the quantitative aesthetic convolutional neural network model are accurate, and further shows that the quantitative aesthetic entropy evaluation model is reliable.

[0121] Table 7 Output results of the intelligent screening program of the measurement aesthetic convolutional neural network

[0122] project Mido value Sorting Solution 1 0.8037 5 Solution 2 0.8327 1 Solution 3 0.8120 2 Solution 4 0.8049 4 Solution 5 0.7892 8 Solution 6 0.8079 3 Solution 7 0.7956 7 Solution 8 0.7645 9 Solution 9 0.7974 6

[0123] Table 8 Results of the questionnaire survey on automobile styling solutions

[0124]

[0125]

[0126] Table 9 Comparison of the ranking results of the beauty values ​​of automobile styling solutions

[0127]

[0128] Therefore, the present invention proposes an intelligent screening method for automobile styling based on a quantitative aesthetic entropy evaluation model. By combining aesthetic measurement theory, information entropy theory and complexity theory, a quantitative aesthetic entropy evaluation model is constructed, which realizes the scientific quantification and objective evaluation of the beauty of automobile styling, and improves the beauty and efficiency of automobile styling design.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. An intelligent screening method for automobile styling based on a quantitative aesthetic entropy evaluation model, characterized in that: The specific steps are as follows: S1. Car styling image data collection and preprocessing, constructing a car styling image dataset for measuring aesthetic convolutional neural networks and a line drawing atlas for calculating car styling beauty values; S2. Based on the Birkhoff aesthetic measurement theory model, entropy evaluation theory model and complexity theory model, a quantitative aesthetic entropy evaluation model is constructed to calculate the beauty value of automobile styling; S3. Construct a quantitative aesthetic convolutional neural network model based on quantitative aesthetic entropy evaluation; S4. Validation of the convolutional neural network model for quantitative aesthetics.

2. According to claim 1, a method for intelligently screening automobile shapes based on a quantitative aesthetic entropy evaluation model is characterized in that: In S1, the preprocessing operation includes grayscale processing, background clipping batch processing, standardization processing, automobile modeling structure line extraction processing and noise reduction processing.

3. The method for intelligently screening automobile shapes based on the quantitative aesthetic entropy evaluation model according to claim 1 is characterized in that: In S2, the specific calculation steps of the automobile styling beauty value are as follows: S21. Measure the entropy redundancy of the three-dimensional modeling lines of the automobile based on the entropy evaluation model and calculate the aesthetic order of automobile modeling; S22. Calculate the aesthetic complexity of automobile styling based on the complexity theory model; S23. Based on the quantitative aesthetic entropy evaluation model, the automobile styling aesthetic order and the automobile styling aesthetic complexity are comprehensively considered to calculate the automobile styling beauty value.

4. The method for intelligently screening automobile shapes based on the quantitative aesthetic entropy evaluation model according to claim 1 is characterized in that: In S21, the calculation formula of the automobile styling aesthetic order is as follows: Among them, O is the aesthetic order of automobile styling, O h For each view’s morphological order, W h is the weight value for each view, h is the four views of the car shape; The morphological order of each view is O h , and its calculation formula is as follows: Among them, S j is the information entropy of structure line j, W j is the weight value of the modeling structure line, f j (x i ) is the car modeling curve function, P ji is the ratio of the derivative value of point i on the jth structure line to the sum of the derivative values ​​of all points of the mth derivative on the structure line, m is the number of points on the curve with different absolute values ​​of derivatives, k = (ln m) -1 , i is an integer, j is an integer, and n is an integer.

5. The method for intelligently screening automobile shapes based on the quantitative aesthetic entropy evaluation model according to claim 4 is characterized in that: The modeling structure line weight value W j The Likert nine-level scale is used to weight the car styling structure line, and the calculation formula is as follows: Among them, P j is the average score of each modeling structure line, and n is the total number of modeling structure lines in each view.

6. The method for intelligently screening automobile shapes based on the quantitative aesthetic entropy evaluation model according to claim 3 is characterized in that: In S22, the formula of the complexity theory model is as follows: C = r × E × t; Among them, C is the aesthetic complexity of automobile styling, r is the simplest order of automobile styling, E is the dissipation of automobile styling information energy, and t is the slow growth rate.

7. The method for intelligently screening automobile shapes based on the quantitative aesthetic entropy evaluation model according to claim 3 is characterized in that: In S23, the quantitative aesthetic entropy evaluation model formula is as follows: Among them, M is the beauty value of the car's styling.

8. The method for intelligently screening automobile shapes based on the quantitative aesthetic entropy evaluation model according to claim 1 is characterized in that: In S3, the quantitative aesthetic convolutional neural network model uses the pre-trained ResNet50 network as the basic model and the mean square error MSE as the loss function.

9. The method for intelligently screening automobile shapes based on the quantitative aesthetic entropy evaluation model according to claim 8 is characterized in that: The calculation formula of the loss function is as follows: Where n is the total number of samples, y λ 、M λ is the true label data of the beauty value of the λth sample, The beauty value prediction data for the λth sample.

10. The method for intelligently screening automobile shapes based on the quantitative aesthetic entropy evaluation model according to claim 8, characterized in that: The quantitative aesthetic convolutional neural network model uses the Adam optimizer to update model parameters, and the model training parameters include the number of iterations epoch, the batch size batch_size and the learning rate.

Images