Research On Target Detection Algorithm Based On YOLO

Target detection is a hot and popular research direction in the field of computer vision today and has important research significance in security monitoring,autonomous driving,drone tracking,and medical imaging.In recent years,with the development of computer software and hardware technology and the emergence of deep learning,target detection algorithms based on deep learning have been continuously proposed,which gradually replaced traditional target detection algorithms based on manual features.YOLO algorithm is an advanced real-time object detection method.Compared with other real-time object detection methods,YOLO algorithm has fast detection speed and good real-time detection effect,but it also has corresponding shortcomings,that is,detection accuracy is not high and small target detection not effectively.Therefore,this article improves and optimizes its algorithm for these problems.The main work is as follows:(1)Aiming at the problem of low detection accuracy of YOLOv3 algorithm,a target detection algorithm combining GIo Uloss and Focal loss is proposed.It is mainly a redesign of the loss function.The loss function of the YOLOv3 algorithm is composed of three parts.This article mainly redesigns the bounding box loss and confidence loss.First of all,the Io U function cannot reflect the problem when there is no intersection between the predicted frame and the real frame and when the predicted frame and the real frame overlap.Compared with the GIo U function,GIo U has the characteristics of IOU and can solve the problem that IOU cannot handle;Secondly,in view of the problem of imbalance between positive and negative samples,the focus loss is proposed as a confidence loss;finally,through the design of the loss function,its feasibility is verified on the PASCAL VOC data set and MS COCO data set.Experimental results show that the improved loss function can improve the detection accuracy.(2)Aiming at the problem that the YOLOv3 algorithm has a higher rate of missed detection of small targets than large targets,an improved YOLOv3 target detection algorithm is proposed,which is mainly to improve the feature fusion in the YOLOv3 network,that is,through the improvement of the feature pyramid and the introduction of an attention mechanism.First,up-sampling and down-sampling are used to fuse the feature maps of various scales,and the fused feature maps are added through the Concat operation;secondly,the attention mechanism module is introduced;finally,the improved network is verified on the data set Its feasibility.Experimental results show that the improved YOLOv3 network has a better detection effect on small targets.