Vehicle Recognition based on Deep Convolution Neural Networks

Abstract

This paper realized the automatic recognition of vehicle with the deep convolution neural network (DCNN) based on deep learning framework CAFFE and GPU which has the strong computing power. And we training the DCNN with 240 class vehicle images of various environments which are obtained at all junctions in a city. After repeated tests, a deep neural network with 13 layers was selected as the feature extraction model. we introduced the super-resolution(SR) algorithm based on deep learning and the illumination compensation algorithm to preprocess the vehicle images so as to improve the quality of the images and extract more robust features. Finally, the experimental results show that the features of the vehicle are more obvious after adding the algorithm of super-resolution reconstruction and light compensation, and the recognition accuracy has been significantly improved.