Research on Vehicle Classification Method Based on Improved AlexNet

Abstract

In the actual traffic intersection environment, vehicles adopted for traffic monitoring miss detection due to various reasons, and because the difference between different types of vehicles is very small, the traditional method cannot effectively distinguish the types of vehicles, and the deep learning image processing method can be used for automatic recognition of vehicle types. We propose an improved AlexNet[1] (ProAlexNet) intersection vehicle classification method by improving and reconstructing the hierarchy and parameters of the AlexNet convolutional neural network structure. In addition, we used a self-made high-quality vehicle category data set, which included 5000 pictures of cars, trucks, buses, motorcycles, and vans. In the experiment, we carried out comparative experiments on three indicators between ProAlexNet network and traditional AlexNet method, and carried out comparative experiments on three traditional recognition algorithms of ProAlexNet. Experimental results show that our improved algorithm has strong competitiveness.