The Research Of Vehicle Classification Based On Deep Learning

The intelligent transportation system has drawn more and more attention with the increase of complex traffic problems.Among them,the vehicle recognition is an important issue in the study of intelligent transportation system.At present,with the development of image processing,pattern recognition,and computer vision,image recognition technology based on deep learning becomes mature and provides a new idea to the problem of the vehicle recognition.Deep learning can independently study features of object,which can avoid artificial features.So,it has a broad application prospect in field of vehicle recognition.This paper is based on the technology plan projects of Hebei Academy of Sciences,and the name of project is “vehicle recognition system based on deep learning”.This paper mainly studies the technology of vehicle recognition based on deep learning.First,this paper summarizes the structure characteristics and training methods of the convolution neural network in the deep learning.According to requirements of the project,this paper proposes a model of vehicle recognition system based on deep learning and discusses the module of vehicle recognition in detail.Then,in view of the current problem that lacks of the public vehicle database,we build a database of vehicle recognition for training and testing models through orientation,induction and annotations.After that,we train the network model with the data.Meanwhile,we constantly adjust and optimize the network structure and form the network model of vehicle fine recognition base on the convolution of the neural network.Finally,we train and test the vehicle images with the help of Caffe deep learning framework and the GPU workstations which equipped with high-performance computing card.The accuracy reaches 97% on the vehicles database which has 164 classes.The experimental results show that the approach of deep learning used in the vehicle fine recognition has a good advantage.