A Research Of Vehicle Recognition Algorithm Based On Deep Learning

Target recognition is very popular in the two fields of computer vision and digital image processing.It is very realistic to use the characteristics of computer vision to reduce labor costs.Object recognition is an important branch of computer vision and an important tool in image processing.Moreover,deep learning is very popular recently,which makes the development of target recognition algorithms based on deep learning very fast.This article has carried out relevant research on the vehicle recognition algorithm based on deep learning,the main work and innovative carrying case are in the following aspects:1.This article has studied the relevant literature data of the classic target recognition algorithm and conducted experiments,and finally chose to use the Faster R-CNN algorithm for improvement.In the initial training process,the FPN network is used to replace the original backbone network.The experimental results show that the recognition accuracy of the FPN network is improved compared with the original network,but the recognition effect of the occluded target is not very ideal.2.This paper introduces the context extraction module and the attention guidance module to the backbone network,which improves the recognition of the occluded target by the FPN network and further improves the recognition accuracy.These two modules help to extract the context information in the multi-layer convolutional neural network and eliminate the redundant information,which helps to capture the information of those unobvious objects,such as occluded targets.The experimental results show that the FPN network with the addition of the context extraction module and the attention guidance module has improved the recognition effect of the occluded target and the overall recognition rate of the algorithm.3.This article uses a method based on key points to improve the Faster R-CNN algorithm,which solves the problem that the algorithm requires a large number of a priori boxes for identification,and a large number of a priori boxes have only a small part of the actual position of the target.The overlap is relatively large.The problem of the imbalance of the ratio of positive and negative a priori frames brought by it also reduces the training and recognition speed of the network.This method only needs to find two points in the upper left corner and the lower right corner to accurately locate a target,because the corner points are easier to extract,so the speed is faster.The experimental results show that the use of the Faster R-CNN algorithm based on the key point method improves the recognition speed better than the original algorithm.