Research On Vehicle Type Recognition Algorithm Based On Deep Learning

With the continuous development of the economy and the improvement of people's living standards,the number of car ownership has shown a rapid growth trend.The increase in the number of cars has changed the way people travel.At the same time,the travel environment has become more complicated and various types of traffic problems have emerged in an endless stream.Cases such as accidents escaping,stealing vehicles,and applying license plates occur from time to time.The method of combining vehicle type and license plate recognition can quickly and effectively identify problem vehicles.This method is of great significance in the detection of the above cases.At present,the license plate recognition technology is relatively mature,and the vehicle identification technology mainly includes the vehicle detection technology and the vehicle classification technology.At present,it is still not perfect.This article originated from one of the pre-research topics of the audit traffic control module of Hisense Network Technology “Transportation Intelligence Management Platform”,focusing on the application of deep learning in vehicle identification.The main work is as follows:The status quo of research on vehicle detection at home and abroad and the status of vehicle classification research are summarized.The development of regional-based convolutional neural networks is introduced and compared.The shortage of the Fasterrcnn method is improved.This paper improves from the following two aspects: optimal sharing The convolutional layer and the introduction of the candidate box voting mechanism;Structurally comparative analysis of the characteristics of different types of convolutional neural networks,using CIFAR-10 open data sets for experiments to verify the classification ability of different convolutional neural networks,to determine the optimal vehicle classification Network;vehicle detection experiment on public data set BIT-Vehicle ID.Three different convolutional neural networks,ZFNet,VGG16 and ResNet,are selected as the improved constellation of Faster-rcnn.After analyzing the experimental results,the detection results of the detection network with ResNet as the shared convolutional layer are the best.The mAP is as high as 94%.In addition,the migration test of the CompCars dataset also shows a good generalization capability.Finally,a vehicle classification experiment is conducted to output the predicted vehicle type of the target vehicle.