On Vehicle Type Recognition Algorithm Based On Deep Learning

With the development of the economy,the number of cars has shown a rapid growth trend,and the intelligent transportation system has received more and more attention,and becoming a very important research focus of the intelligent traffic management system.As an important branch and key technology of intelligent transportation system,it has a widely application in the fields of intelligent parking toll collection system and road traffic condition supervision,so vehicle identification has become one of the research hot spots of scholars at domestic and foreign in recent years.However,due to the late start of China’s vehicle identification technology,there is still a certain gap between the technology and the developed countries,so the intelligent transportation system has not been fully popularized in China.In this paper,the characteristics of the existing vehicle identification algorithm are complex,time-consuming,and low recognition rate.The vehicle identification algorithm is studied by comparing various excellent network models and algorithms.The main work is as follows:First of all,this paper studies and analyzes the traditional vehicle identification technology based on traditional machine learning,and compares the advantages and disadvantages of traditionally used feature operators and classifiers.Then for the deficiencies of the traditional methods,a vehicle classification method based on convolution neural network is proposed.After analyzing and comparing various network models and optimization algorithms,we adopts the third generation network model in GooLeNet,and it is equipped with a series of excellent algorithms such as Adam and Dropout.In order to improve the recognition accuracy of the network,the AM-Softmax loss function is proposed to replace the original Softmax loss function,and the model is pre-trained by the migration learning idea.Finally,the data enhancement strategy is used to amplify the data set to improve the accuracy of the model.After simulation verification on the BIT-Vehicle data set,the results show that the designed network model recognition accuracy is better than the existing methods.Secondly,we introduced the two mainstream target detection algorithms,and compared the advantages and disadvantages of the target detection algorithm based on regional recommendations and the regression-based target detection algorithms.By combining the specific research content of this paper,the latest network model in the YOLO series with faster detection speed is adopted,and various excellent optimization algorithms are used to improve network performance.After learning the more mature technology in face recognition,the latest Arcface loss function in face recognition is used to replace the original binary cross entropy loss function,which further improves the overall performance of the model.The experimental results show that the improved network model recognition accuracy has been further improved.Finally,the main work done in this paper is summarized,and the research direction of future vehicle identification technology is prospected.