A Big-data Driven Automatic Evaluation Method For Automobile Exterior Styling

The degree to which a car’s styling is accepted and loved by consumers can directly affect the competitiveness in the market.However,before the car has not been tested in the market,it is difficult to accurately judge the preference of potential users due to the interference of many subjective and objective factors.Therefore,constructing an objective and reasonable automatic evaluation method for automobile styling has important research value for automobile styling design.To solve this problem,this paper takes the automatic evaluation of automobile styling as the research objective,and carries out research on the comprehensive scoring mechanism of automobile styling considering user attributes,the automatic evaluation mechanism of automobile styling driven by big data and some other related series of problems.The specific research work is as follows:In order to improve the accuracy and rationality of the automatic evaluation of automobile styling,an accurate recognition and angle estimation method of automobile models based on Res Net-50 feature extraction network is proposed.The network collected and created a largescale automobile multi-view uniformly sampled image data set USMV875 including 875 models and 54750 images,which provides data support for subsequent model identification and angle estimation.It is also used for vehicle model recognition and angle regression estimation.Numerical experiments show that the accuracy of the car model recognition model in this paper can reach 98.61%,and its performance is better than Alex Net,VGGNet-16 and Res Net-18 network structures.The angle estimation error of the automobile angle estimation model is not more than 3 degrees,which has good accuracy and robustness.In addition,this paper explores the impact of different angle areas in the training set on the recognition rate of car models,and provides a reference for the angle factors of the automatic evaluation mechanism.This paper proposes an automatic evaluation mechanism based on deep learning for automobile styling users.And it collects and establishes the user evaluation database of automobile styling based on the representation of multi-view images.Considering the users’ attribute information,this paper constructs a comprehensive scoring mechanism for automobile styling to ensure the impartiality and professionalism of user scoring,and then process the evaluation data of network users to complete the styling score labeling of automobile models.The regression method based on the Res Net-50 feature extraction network is used to train the car styling scoring model and analyze the influence of angle on the car styling score to construct an automatic user evaluation machine for car styling combining the car model recognition model and the angle estimation model.Experiments show that this method can obtain a more objective and reasonable evaluation of automobile styling,which has certain reference value for designers to grasp user preferences.A DRGAN-based generative adversarial network is proposed to realize the automatic generation of multi-view images of automobiles.This paper adopts the DRGAN method,considers the actual data situation,and uses the multi-view car model image training model of each model under the USMV275 data set to realize the generation of the car’s multi-view image.Numerical experiments show that inputting different angles of car images,the angular distance between the generated image and the input image,and different car brands will all have an impact on the effect of the generated image.