Research On Vehicle Detection Methods Based On Deep Learning

With the rapid growth of the number of motor vehicles,the urban transportation system is facing severe challenges.Intelligent transportation systems have played an important role in improving transportation efficiency and reducing hidden dangers in traffic safety,and have received widespread attention.Vehicle detection technology is one of the basic problems of intelligent transportation systems.With the advent of deep learning,vehicle detection algorithms based on convolutional neural networks have been further developed.However,the performance of vehicle detection algorithms needs to be improved due to the complex and changeable scenes in real traffic,various vehicle sizes,lighting changes,and occlusion.This article focuses on vehicle detection methods based on deep learning.The specific research content is as follows:Firstly,in order to improve the classification accuracy of vehicle detection,this paper propose a vehicle detection network based on the attention mechanism.This method integrates the channel and spatial attention modules into the backbone network,which effectively helps the information flow in the network and improves the feature extraction capability.Experimental results show that the proposed method significantly improves the classification accuracy of vehicle detection without introducing additional calculations.Secondly,in order to improve the regression accuracy of vehicle detection,this paper propose a vehicle detection network based on a selective feature fusion strategy.This method uses a simple and effective selective fusion method to reorganize the output features at different levels in the feature pyramid,without compromising the semantic representation of high-level features,and enriching its position information.Experimental results show that the proposed method not only reduces the classification error of vehicle detection,but also improves the regression accuracy of vehicle detection.Finally,in order to improve vehicle detection efficiency,this paper propose a vehicle detection network based on network pruning strategy.This method uses the₁l norm to constrain the proportionality coefficient in the network batch normalization layer,so that the model adjusts the parameters in the direction of sparse structure.Network pruning is accomplished through a proportionality factor and a predetermined pruning rate.Experimental results show that the proposed method achieves a significant improvement in detection efficiency within the allowable range of detection performance attenuation.