摘要
为了准确快速地筛选捕捞鲜对虾中破损虾以及贝类等杂质,将基于Pytorch框架的YOLO网络应用于自主研发的海产品智能分拣装备,进行对虾识别速度和精度的研究,自主创建多样性对虾数据集并进行数据增强,使训练模型具有更高的鲁棒性。YOLO v4、YOLO v5s的查准率、召回率、F1评分、多类别平均精度分别为94.02%、95.22%、0.95、95.67%和90.26%、90.47%、0.91、90.67%。YOLO v4、YOLO v5s平均个体的检测速度分别为47.22、30.32 ms。
To accurately and quickly screen freshly caught prawns for impurities such as broken prawns and shellfish,the YOLO network based on the Pytorch framework was applied to the self-developed seafood intelligent sorting equipment for the study of the speed and accuracy of prawn identification.Diverse prawn datasets were autonomously created,enhancing the robustness of training models.The precision,recall,F1 score and mean average precision values of YOLO v4 and YOLO v5 s were 94.02%,95.22%,0.95,95.67%;90.26%,90.47%,0.91,90.67%,respectively.The average individual detection speed was 47.22 and 30.32 ms.
作者
付凯月
冯怡然
陶学恒
FU Kaiyue;FENG Yiran;TAO Xueheng(School of Mechanical Engineering and Automation,Dalian Polytechnic University,Dalian 116034,China;National Marine Food Engineering Technology Research Center,Dalian Polytechnic University,Dalian 116034,China)
出处
《大连工业大学学报》
CAS
北大核心
2022年第5期379-385,共7页
Journal of Dalian Polytechnic University
基金
辽宁省自然科学基金项目(2020-MS-273)
大连市青年科技之星项目支持计划项目(2021RQ088)。