A fine-grained recognition method for car models from multiple angles

Abstract

The invention discloses a multi-angle fine identification method for vehicle models. In view of the influence of external environmental factors, the present invention adopts the SSD vehicle target detection framework based on ResNet-50 to detect and locate the vehicle target in the image, and reduce the interference of external environmental factors on vehicle identification; for the scale deformation of vehicles under the condition of multi-view problem, the present invention performs multi-scale fusion of the features of different convolutional layers of B-CNN, improves the feature expression ability, and introduces the center loss, constrains the vehicle image in the feature space, and effectively guides the network learning so that the class Features with small distance and large inter-class distance.

Description

technical field [0001] The invention belongs to the technical field of image classification, and in particular relates to a vehicle image classification method. Background technique [0002] The automobile is one of the most important inventions in the history of modern human civilization. With the rapid development of my country's social economy, the continuous expansion of population and the acceleration of urbanization, the number of automobiles in my country is also increasing. On July 6, 2018, data released by the Traffic Management Bureau of the Ministry of Public Security showed that as of the end of June, the number of motor vehicles in the country reached 319 million, of which the number of private cars reached 180 million, and maintained a continuous and rapid growth trend. While the number of cars continues to increase, many traffic problems have also emerged and become increasingly prominent, such as the problem of licensed cars, illegal parking, illegal U-turns,...

AI Technical Summary

Problems solved by technology

[0005] 1. In the real scene, not only the appearance of the vehicle itself is complex and changeable, but also the influence of external environmental factors such as lighting and background in different scenes further increases the difficulty of vehicle identification

[0006] 2. In the case of multiple viewing angles, the vehicle has a certain scale change and deformation, and it is difficult to find effective and stable visual features for accurate recognition

In addition, due to different shooting angles of the same type of car, the intra-class distance may be greater than the class distance, which may cause the vehicle model to be misidentified

Images