The Research Of Vehicle Model Fine-grained Recognition Method Based On Attention Model

Vehicle is very important for people’s daily travel activities,and it is also one of the main research objects of intelligent transportation system.In recent years,the number of vehicles has increased dramatically,which not only brings great pressure to the work of traffic management departments,but also poses greater challenges to the development of intelligent transportation technology,in which the task of vehicle model recognition is very important.However,there are a large number of vehicle models,some models have small appearance differences,and the actual shooting environment is complex.These factors greatly increase the difficulty of fine-grained vehicle model recognition.Aiming at the research of vehicle model recognition task,this paper combines visual attention mechanism with deep learning,and proposes two kinds of vehicle model fine-grained recognition methods from multiple research points,namely,vehicle model fine-grained recognition based on dual-rank attention feature fusion and a semantic enhanced 3D attention network for vehicle model fine-grained recognition.The experimental results show that our attention model can effectively improve the recognition effect of vehicle model.The main work of this paper is as follows:(1)A vehicle fine-grained recognition method based on dual-rank residual attention fusion module is proposed: The convolutional neural network is used as the main baseline,and the spatial transformer network is used as the attention coder to weight the global attention of the original vehicle image.Then,multiple features of different levels in the network are collected and fused to enhance the richness of classification feature information,and the channel attention structure is embedded to enhance the feature response of the current category.(2)A semantic enhanced 3D attention network for vehicle fine-grained recognition is proposed: The algorithm consists of semantic enhancement module and3 D attention module.The semantic enhancement module is used to improve the expression ability of attention mechanism in the network,and the 3D attention is used to mine the discriminative features of vehicles.The 3D attention module is compared with the original baseline network on the vehicle datasets,which proves the effectiveness of the 3D attention module for vehicle recognition task.