Research On Target Detection And Recognition Algorithm Based On Deep Learning

The detection and identification of road traffic targets is a key link of the Intelligent Transport System(ITS).It can locate the specific position of vehicles and pedestrian target in the image and output the category of each target,which is of great significance for studying intelligent transportation.Traditional target detection and recognition methods have achieved good results,but they are susceptible to background and light changes in light and dark.These methods have the disadvantage of low recognition accuracy.This paper mainly studies the target detection and recognition algorithms based on deep learning,and focuses on solving the problems of difficult to distinguish similar vehicles,insufficient use of effective image information,and difficult detection of small targets.The main contents are as follows(1)Aiming at the problem that it is difficult to classify similar vehicles more precisely and accurately,an algorithm for small vehicle target detection and recognition based on residual network is proposed.The algorithm changes the composition mode of the original convolutional neural network to a residual mode based on local connection and weight sharing,and changes the network model to reduce complexity.The low-level features extracted from the front layer of the image and the high-level features extracted from the back layer are fused and combined,and the region of interest pooling method is used to unify them into a feature matrix of the same size.Finally,the confidence and correction parameters of the target bounding box is obtained through the classification layer and the regression layer.Experiments show that the improved model can enhance the learning ability of the network and improve the average detection accuracy under the premise of ensuring time efficiency.This method has achieved good results on the detection of similar small vehicles.(2)Aiming at the problems of excessive redundant information,insufficient effective information,and difficulty in accurately identifying distant and blurry targets in daily road traffic images,the CBAM(Convolutional Block Attention Module)is first introduced into the traffic target detection and recognition algorithm.The attention mechanism combines a channel and a spatial attention mechanism.The channel attention focuses on describing what the target is and the spatial attention focuses on describing where the target is.The two methods combine to enhances the ability to express features.Compared with the traditional single attention model,this method has better detection and recognition effect.Secondly,a two branch Rol-Pooling method is used to improve the algorithm,using the multi-layer features of the image.The candidate region mapping process is divided into two ways,which are respectively mapped to feature maps containing coarse and fine-grained features.Then the two branches are merged to improve the ability to portray the details of the image,thereby improving the detection effect of road targets.Finally,the above two methods are used to improve the Faster-RCNN algorithm,and experiments are designed from four different angles to verify the effectiveness of the algorithm.Experimental results show that the improved model has achieved good results in road traffic target detection and recognition.