基于YOLOv5算法的舍饲羊只检测与跟踪方法研究

打开文本图片集

关键词：多目标检测与跟踪；深度学习；舍饲羊只；遮挡粘连；ID跳变；精准养殖中图分类号：S826；TP391.4 文献标识码：A 文章编号：2095-5553（2025）12-0070-07

Abstract：Precision farming is aneffective way to improvethequalityandefficiencyof modern sheep farming，in which the intelligent detectionandtrackingofsheep isakeylink intheimplementationofprecision sheepfarming.However， theflockcharacteristicsof sheep make the phenomena of omission，misdetectionandID switching prominent when implementing multi-target tracking due to the occluding and adhering between sheep.Inview of this，this paper uses diferent detection targetanchoringmethodsandapplies the YOLO+DeepSortalgorithm toimplement multi-target detection and tracking of sheep in a herd environment，andYOLOv5+DeepSort andYOLOv8+DeepSort algorithms are usedtomakethecontroltest.Theresults show thatthedetection and tracking acuracy and ID switchesof the improved method are improved，in which the tracking accuracy based on YOLOv5+DeepSort algorithm can reach 96.73% ，which is 3.22% higher thantheoriginal method，and the number of ID switches isreduced by18 times，andthe tracking accuracy of YOLOv8+DeepSort algorithm can reach 94.85% ，which is 13.59% higher than the original method，and the number of IDswitches isreducedby9 times.The improved anchoring method namely improves the individual recognition accuracyoftheflock sheep，andatthe sametimereduces the interferenceinformation inthe individual index of thetarget object，whichisconduciveto further information extraction.

Keywords：multi-target detection and tracking；deep learning；housed sheep；occlusion adhesion；ID switches; precision farming

0 引言

内蒙古作为我国最大的畜牧业基地之一，养羊业是重要的经济产业，对于农村地区的经济增长和创造就业起到了重要作用1。（剩余12177字）