Research On Vehicle Recognition Algorithm Based On Deep Learning

Vehicle type recognition is an important technology in modern intelligent transportation.The existing deep convolution network has achieved good results,but the current categories mainly stay in the coarse-grained classification of the general shape of vehicles,while the research on the fine classification of vehicles is less,and the accuracy of classification and operation time also need to be further improved.This paper studies the correct classification of vehicle brand and model to improve the accuracy of fine-grained vehicle identification.This paper studies the vehicle recognition method based on deep learning,selects the deep residual network as the basic classification network and improves it,introduces the second convolution block into the dense connection,strengthens the characteristic flow of different depth convolution layers in the network,so as to improve the characteristic expression ability of the network;at the same time,introduces the random weight average and label smoothing into the residual network,on the one hand,makes On the other hand,the generalization ability of the model is strengthened to ensure the accuracy of fine-grained classification.In view of the situation that different brands of vehicles in vehicle model recognition are identified incorrectly due to the similarity of vehicle models,a new multi-scale residual model is designed based on the RESNET network model,which increases the width and depth of the network,accelerates the convergence of the network at the same time when the network uses the high computing performance of the dense matrix,so as to achieve the purpose of enhancing the feature extraction of the network.A new loss function is proposed and used to design a thick and thin double label.Through the combination of the thick label and the thin label,the recognition accuracy is further improved.In this paper,we use the knowledge learned from a large number of data in the Imagenet database to optimize the network parameters through the migration learning method.We mark the Stanford cras-196 data set in a fine-grained way,divide it into training set and test set for model training and learning,and finally extract the effective features for vehicle classification.The accuracy of the two improved networks is 83.7%and 87.4%respectively.The experimental results show that the method proposed in this paper has a high accuracy,which verifies the effectiveness of the method.