Research On Vehicle Target Detection Method Based On Deep Learning

With the rapid development of economy,car ownership continues to rise.While cars bring convenience to people,they also bring a series of problems such as traffic congestion.In recent years,intelligent transportation is in the ascendant,so higher requirements are put forward for intelligent transportation system.Vehicle detection technology is the basis of intelligent transportation system and plays an important role in its development.In the process of vehicle detection,due to the change of camera angle and distance between vehicle and camera during image acquisition,some target vehicles to be detected are too small,resulting in poor detection effect and serious missing and misdetection phenomenon.Therefore,this paper conducts in-depth research from the dimension of feature channels and proposes an AFFNet（Attention Feature Fusion Network）model,which optimizes the feature fusion method and maximizes the "accuracy" and "diversity" of features and reduces the the feature maps to be detected.This paper applied AFFNet to the YOLOv5 detection model and achieved good detection results,enabling end-to-end vehicle detection.The main work of this paper is as follows:1.This paper proposes a feature attention fusion network model AFFNet,which introduces a spatial attention fusion module SAF and a multi-layer feature fusion module MFF.SAF module generates spatial attention through attention operation.The MFF module divides the feature channels,making some channels pay attention to the detailed features of the object and the other to the deep semantic information of the object.In AFFNet,three SAF modules are used to generate different spatial attention,and one MFF module is used to distinguish the importance of channels,so as to enhance the semantics and detail features in the fusion stage as a whole.2.AFFNet model is applied to YOLOv5 detection model,which maximizes the "accuracy" and "diversity" of features,reduces the feature maps to be detected,improves the detection accuracy,and implements end-to-end vehicle detection,which has certain practicability.3.Swin Transformer and its application in vehicle detection are studied.This paper introduces the whole structure of Swin Transformer and its shfit window strategy,which is applied to vehicle object detection and achieves good results.