摘要
为提高马铃薯种薯芽眼的识别效果,采用一种端到端的YOLOv3目标检测算法对马铃薯种薯芽眼进行识别。通过调整试验过程中的超参数学习率、批次和迭代次数进行训练和测试马铃薯图像数据集。实验结果表明:YOLOv3算法对马铃薯表面良好的种薯识别精确率P为92.18%、召回率R为91.66%,测试速度为42FPS;对种薯表面含有泥土、机械损伤等状况进行检测,种薯芽眼识别平均精度为86.54%。YOLOv3模型对种薯芽眼的识别效果保持了较好的实时性要求,可为马铃薯种薯自动化切块中的种薯芽眼识别提供参考。
In order to improve the recognition effect of potato seed tuber sprout eyes,an end-to-end YOLOv3 target detection algorithm is used to identify potato seed tuber sprout eyes.The potato image data set is trained and tested by adjusting the hyperparameter learning rate,batches and iteration times during the experiment.The experimental test results show that the YOLOv3 algorithm has a good seed potato recognition accuracy P of 92.18%,a recall rate R of 91.66%,and a test speed of 42FPS.The detection of soil and mechanical damage on the surface of seed tuber showed that the average accuracy of seed tube sprout eye detection was 86.54%.The YOLOv3 model can maintain good real-time requirements for the recognition effect of seed tuber sprout eyes,and can provide a reference for seed tuber sprout eyes recognition in automatic cutting of potato seed tuber.
作者
陈志伟
张万枝
张涛
徐颖
吕钊钦
类成龙
Chen Zhiwei;Zhang Wanzhi;Zhang Tao;Xu Ying;Lv Zhaoqin;Lei Chenglong(College of Mechanical and Electronic Engineering,Shandong Agricultural University,Tai’an 271018,China;Nanjing Research Institute of Agricultural Mechanization,Ministry of Agriculture and Rural Affairs,Nanjing 210014,China;Tengzhou Agricultural and Rural Comprehensive Service Center,Zaozhuang 277500,China)
出处
《农机化研究》
北大核心
2022年第11期19-23,30,共6页
Journal of Agricultural Mechanization Research
基金
中国博士后科学基金项目(2020M681690)
山东省自然科学基金项目(ZR2019BC018)
山东省农业重大应用技术创新项目(SD2019NJ003)
山东省现代农业产业技术体系薯类创新团队农业机械岗位专家项目(SDAIT-16-10)
“十三五”国家重点研发计划智能农机装备专项(2017YFD0700705)。
作者简介
陈志伟(1997-),男,安徽宿州人,硕士研究生,(E-mail)2651932902@qq.com;通讯作者:张万枝(1986-),男,山东枣庄人,博士后,硕士生导师,(E-mail)zhangwanzhi@163.com。
