Vehicle Target Detection And Recognition With Attention Mechanism

With the rapid development of modern science and technology,intelligent and digital vehicle detection technology has received unprecedented attention.In the past ten years,image target detection algorithms based on depth learning have made breakthrough progress.It has robustness in complex scenes.In this paper,a vehicle target detection system is designed to detect and track the target accurately,efficiently and automatically.In this paper,the target detection algorithm is studied and analyzed,and the Faster R-CNN(Faster Region-Based Convolutional Neural Networks)model with better detection accuracy is used as the foundation.Multi-scale and attention mechanism are added to improve it,and the experimental verification is carried out.The main work and improvement of this paper are as follows:1.Collectting and expanding the vehicle data set of complex scenes,and data enhancement and human intervention to expand the data set.Different vehicle detection model algorithms are tested and their principles are compared and analyzed.Considering that different algorithms adapt to different datasets,the data sets at home and abroad and the data obtained after expansion are combined and trained.Finally,the real-time detection effects of the algorithm models are compared.2.Attention mechanism and multi-scale are added to make the algorithm suitable for more complex scenes.The model is extracted with multi-scale features,and then the image feature information fusion is predicted by inference,so the anti-interference ability of the algorithm model in vehicle detection can be improved.Considering semantic features such as time,attention mechanism is added to mine powerful hidden image information.The improved algorithm can mine deep feature information and hidden information during real-time reasoning,thus the detection accuracy of the network is improved.The experimental results show that the performance index of the improved algorithm is higher in different data sets,and thedetection accuracy is 2.6% higher than the original algorithm.