Super-Resolution Image Reconstruction And Vehicle Identification Based On Deep Learning

In view of the increasingly serious traffic problems,people try to improve by intelligent transportation system.The rapid development of deep learning technology provides a new solution for the related problems in the field of intelligent transportation.This paper studies the image reconstruction and vehicle type recognition in the intelligent transportation system based on the deep learning technology.At first,this paper deeply studied the convolutional neural network,the depth of the residual network and several typical convolutional neural network model,analyzes the existing algorithms in the reconstruction of the insufficiency of the effect,speed and recognition rate,improved on the basis of this,puts forward the image super-resolution reconstruction based on Laplace structure and the improved residual wide network of vehicle recognition method.LapMSRN is a reconstruction model based on Laplacian structure.However,it uses the same convolution function in each layer,so it is inevitable to lose some information in the up-sampling process.To solve this problem,this paper introduces multi-channel mapping to extract more abundant features,and uses convolution cascade and weight sharing to reconstruct the image in super-resolution.At the same time,in order to optimize the whole network,this paper chooses the MSRA initialization method to initialize the network weight,so as to speed up the model convergence.The new training strategy and activation function PReLU are adopted to shorten the training time and enhance the system stability.Experiments show that the improved model can reconstruct the texture and detail of the image better.In order to meet the needs of vehicle type identification in the real scene,an improved wide residual structure network model is proposed in this paper.The network widens and improves its structure,uses fewer parameters to extract features,and speeds up the convergence of the whole network.The learning method based on A-Softmax function was used to increase the distance between classes and reduce the distance between classes.Experiments show that this method can recognize vehicle types in complex traffic scenes such as intersection,avenue,park road,etc.