A CNN based car model recognition improvement

Abstract

Car model recognition plays an important role in the road auto tolling system and the traffic volume statistics. In this paper, we present an improvement on car model recognition based on convolutional neural network. Firstly, our car image database which includes train and test images has been established. Then we preprocess our car model images using Gaussian Blur. Finally, we choose the appropriate CNN network structure Caffenet and improve the structure of it by adding an additional layer. The results indicate that through our series of improvement strategy, recognition accuracy has been improved to a better degree in terms of our database.