Research On Vehicle Re-Identification Method Based On Deep Learning

With the wide application of deep learning in the field of computer vision,the performance of vehicle re-identification algorithms based on deep learning has been greatly optimized,and it has been successfully applied to traffic safety,intelligent transportation and criminal investigation and other fields.Although the emergence of deep learning has promoted the development of the field of heavy recognition,it is still difficult to extract vehicle features due to complex scenarios and changeable external factors.Therefore,this paper aims to analyze and research around the vehicle re-identification algorithm based on deep learning.Firstly,in order to improve the performance of the vehicle re-identification model,a vehicle re-identification method based on the improved Transformer is designed.Because when Transformer extracts features,it pays more attention to global features and ignores local feature information.Therefore,a method is designed to effectively use global and local features.In addition,because the shooting angle and the distance between the camera and the vehicle have been changing,resulting in the performance of the vehicle re-identification model degrading,this paper integrates non-visual information as auxiliary information into the extracted vehicle features.Secondly,in order to further improve the performance of the vehicle re-identification model,a multi-scale vehicle re-identification method based on Swin Transformer is proposed.The feature information of different scales of vehicles is extracted by multi-scale methods,and the features of different scales are fused to achieve complementary advantages of multi-scale feature information.In order to further extract global feature information and edge detail information from the fused features,a feature enhancement module is designed,which uses the residual fast Fourier transform to realize the interactive feature extraction between frequency domain and spatial domain.Finally,in order to solve the influence of viewpoint variability on the vehicle reidentification task,a vehicle re-identification method based on the viewpoint perception attention mechanism is proposed.The feature proofreading of the channel attention mechanism is used to enhance the characteristic information of the perspective obtained in the same direction,and the feature information of different directions is suppressed.In addition,the local feature learning module under perspective perception is designed,which uses the method of erasing and cropping to process the obtained local feature information to mine more feature information in the corresponding perspective features.