基于YOLOv 4的机收甜菜破损检测方法研究  

Breakage Detection Method of Mechanized Harvesting Beets Based on YOLOv4

作　　者：孙志坚 王健 时佳[1,2] 白晓平 Sun Zhijian;Wang Jian;Shi Jia;Bai Xiaoping(Shenyang Institute of Automation Chinese Academy of Sciences,Shenyang 110000;Institutes for Robotics and Intelligent Manufacturing,Chinese Academy of Sciences,Shenyang 110169,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院沈阳自动化研究所,沈阳110000 [2]中国科学院机器人与智能制造创新研究院,沈阳110169 [3]中国科学院大学,北京100049

出　　处：《农机化研究》2023年第5期6-12,共7页Journal of Agricultural Mechanization Research

基　　金：国家重点研发计划项目(2016YFD0701200);中国科学院科技服务网络计划项目(KFG-STS-QYZD-198)。

摘　　要：针对甜菜机械化收获中机收甜菜识别不精确致使破损率计算不准确的问题,提出一种基于YOLOv4的机收甜菜破损检测方法。利用不同距离、不同角度和不同遮挡程度的甜菜照片制作数据集,对基于YOLOv4的机收甜菜破损检测模型进行训练和测试。测试结果表明:基于YOLOv4的机收甜菜破损检测模型识别完整甜菜精确率和召回率分别为94.02%和91.13%,识别破损甜菜的精确率和召回率分别为96.68%和95.21%,破损检测模型的mAP值为96.44%,比Faster R-CNN和SSD模型的mAP值分别高2.62%和5.65%。由此可得,提出的基于YOLOv4的机收甜菜破损检测模型可以更准确地完成对机收甜菜中完整甜菜和破损甜菜的识别,满足甜菜破损率计算的需求。In order to solve the problem that it is so difficult to detect mechanized harvesting beets accurately and comprehensively that the calculation of breakage rate of mechanized harvesting beets becomes inaccurate.A breakage detection method of mechanized harvesting beets based on YOLOv4 was proposed in this paper.Beet photos from different distances,different angles and different degrees of occlusion were used to make data sets,and the breakage detection model was trained and tested.The test results showed that the accuracy and recall rates which the breakage detection model based on YOLOv4 detects whole beets were 94.02% and 91.13%,the accuracy and recall rates which the breakage detection model based on YOLOv4 detects damaged beets were 96.68% and 95.21%,and the mAP of the breakage detection model based on YOLOv4 was 96.44%.The mAP of the breakage detection model based on YOLOv4 were higher than mAP of the breakage detection model based on Faster R-CNN or SSD.It can be concluded that the breakage detection model based on YOLOv4 proposed in this paper can more accurately detect the whole and damaged beets and meet the requirement of calculating the damage rate of beets.

关 键 词：机械收获 深度学习 YOLOv4 目标识别 甜菜 

分 类 号：S225.72[农业科学—农业机械化工程]