基于YOLOv5-en算法的草莓采摘机器人目标检测技术  

Target detection technology of strawberry picking robot based on YOLOv5-en

作　　者：高云茜 邓三鹏[1] GAO Yunqian;DENG Sanpeng(Robotics and Intelligent Equipment Research Institute,Tianjin Vocational and Technical Normal University,Tianjin 300222,China;School of Mechanical and Electrical Engineering,Heilongjiang Agricultural Economics Vocational College,Mudanjiang 157041,China)

机构地区：[1]天津职业技术师范大学机器人及智能装备研究院,天津300222 [2]黑龙江农业经济职业学院机电工程学院,黑龙江牡丹江157041

出　　处：《实验技术与管理》2023年第10期178-183,216,共7页Experimental Technology and Management

基　　金：天津市科技军民融合重大项目(18ZXJMTG00160);黑龙江农业经济职业学院自然科学类科研项目(NJKY2021015)。

摘　　要：为实现草莓采摘机器人在温室中对草莓的快速、精确检测,该文提出一种YOLOv5-en算法。在原有YOLOv5基础上,首先对原始主干进行网络卷积操作并加入CBAM模块;其次,引入BiFPN模块进行多尺度特征融合;然后,使用直方图均化算法和Mosaic数据增强进行目标检测数据预处理;最后,优化K-means算法对训练数据集聚类分析,同时使用Focal loss损失函数构建YOLOv5-en目标检测网络。通过对比试验得出:与YOLOv5、Faster-RCNN ResNet101和Faster-RCNN VGG16模型相比,YOLOv5-en的mAP分别提升了3.41%、17.85%和14.40%,可达94.36%。通过采摘机器人模拟环境检测验证了该模型的可行性,且达到了应用水平,可为采摘机器人实时小目标检测提供支撑。In order to realize quick and accurate detection of strawberries by strawberry harvesting robot in greenhouse environment,this paper proposes an improved YOLOv5(YOLOv5-en)algorithm.Based on the original YOLOv5,CBAM module is added after convoluting the original backbone network.Secondly,BiFPN module is introduced for multi-scale feature fusion.Then,using histogram equalization algorithm combined with Mosaic data to enhance data pre-processing.Finally,optimized K-means algorithm for clustering analysis of training set and Focal loss function constructed YOLOv5-en target detection network.Through comparative tests,we found that compared with YOLOv5,Faster-RCNN ResNet10 and Faster-RCNN VGG16 models,YOLOv5-en enhances mAP by 3.41%,17.85%,14.40%,reaching 94.36%.The feasibility of the model is verified by the simulation environment detection of the picking robot,and the application level is reached and can provide support for real-time small target detection of harvesting robot.

关 键 词：YOLOv5-en CBAM BiFPN 采摘机器人 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]