Researchon Vehicle Target Detection Algorithm Based On Deep Learning

With the development of our society and the improvement of people’s living standards,vehicles have become an important means of transportation for families,and the pressure of traffic has also increased,and traffic congestion has become more serious.Urban construction and renovation can only solve urgent problems,but cannot effectively solve this social problem for a long time.With the continuous development of computer technology,the intelligent transportation system based on deep learning provides a more reasonable and efficient solution to urban transportation problems.The intelligent transportation system can identify vehicle targets and count the number of vehicles through images and other information,and plan traffic guidance schemes based on the traffic conditions of the entire city based on the number of vehicles to solve the problem of traffic congestion.Therefore,how to accurately and quickly identify vehicle targets from the existing traffic information and accurately count the number,so as to plan traffic routes and solve the problem of urban traffic congestion according to the number of vehicle targets has very important research significance.Aiming at the problem of accurate and rapid identification of vehicle targets and accurate statistics,comprehensively uses deep neural network,residual neural network,hole convolution and other theories and methods to carry out the deep learning-based vehicle target detection algorithm.theoretical analysis and experiments.The main research work is as follows:(1)In order to meet the requirements of accurate and rapid identification of vehicle targets,a vehicle target detection algorithm is proposed which has both the fast single-step detection algorithm and the high accuracy of the double-step detection algorithm.The algorithm first selects a deep convolutional neural network structure to improve the ability to obtain semantic information features,and cooperates with the residual idea to greatly reduce the impact of the disappearance of gradient data during training,and then uses upsampling and downsampling to obtain more Finally,the improved clustering method and anchor point mechanism are involved in the process of classifying and calculating feature information,which can obtain higher detection accuracy and faster detection speed.Through experiments,the effectiveness of the vehicle target detection algorithm in this paper is verified.(2)In order to accurately count the number of vehicle objects,a vehicle object number statistics algorithm fused with atrous convolution is proposed.The algorithm uses the feature pyramid to achieve multi-scale fusion of the detail information in the shallow network and the high-order semantic information in the deep network,and introduces atrous convolution at the front end of the network to increase the receptive field,thereby improving the network performance and accurately counting the number of vehicles.By comparison,the vehicle number statistics algorithm in this paper outperforms other algorithms.