王红军1　邹伟锐1　谢启旋2　郑文和3

（1.华南农业大学，广东 广州 510642   

2.比亚迪股份有限公司，广东 深圳 518118

3.比亚迪电子（国际）有限公司，广东 深圳 518118）

摘要：香蕉采摘是一个耗时费力的作业过程。为了实现香蕉的自动化采摘，首先，设计了一款香蕉智能采摘装置，主要包括叉剪升降机构、3-RPS并联机构和XY直线移动机构等，通过控制这3个机构的运动变化，满足其搭载的末端执行机构适应香蕉果柄的空间随机性要求；然后，搭建双目相机视觉识别系统，采用YOLOv5算法识别香蕉果柄，选择检测框的中心位置为采摘点；接着，根据采摘点位置反求3-RPS并联机构的位姿，驱动切割机构对香蕉果柄定位，完成香蕉串的自动采摘；最后，研制了香蕉智能采摘装置样机，开展采摘试验。试验结果表明，香蕉智能采摘装置的绝对位置误差小于8.66 mm，相对位置误差小于0.85%；γ角度误差小于1.10°，β角度误差小于1.25°，满足香蕉采摘的定位要求。

关键词：机器视觉；YOLOv5；香蕉采摘；3-RPS并联机构；香蕉果柄识别

中图分类号：S225.93; TP391.41     文献标志码：A        文章编号：1674-2605(2023)05-0003-09

DOI：10.3969/j.issn.1674-2605.2023.05.003

Research on Banana Stem Recognition and Picking Experiment 

Based on Machine Vision 

WANG Hongjun1  ZOU Weirui1  XIE Qixuan2  ZHENG Wenhe3

(1.South China Agricultural University, Guangzhou 510642, China 

2.BYD, Co., Ltd., Shenzhen 518118, China 

3.BYD Electronics International Co., Ltd., Shenzhen 518118, China) 

Abstract: Banana picking is a time-consuming and laborious process. In order to achieve automated banana picking, firstly, an intelligent banana picking device was designed, which mainly includes a fork shear lifting mechanism, a 3-RPS parallel mechanism, and an XY linear moving mechanism. By controlling the motion changes of these three mechanisms, the end effector mechanism it is equipped with meets the spatial randomness requirements of the banana handle; Then, build a binocular camera visual recognition system, use YOLOv5 algorithm to identify the banana stem, and select the center position of the detection box as the picking point; Next, reverse the pose of the 3-RPS parallel mechanism based on the picking point position, drive the cutting mechanism to position the banana stem, and complete the automatic picking of the banana string; Finally, a prototype of an intelligent banana picking device was developed and picking experiments were conducted. The experimental results show that the absolute position error of the intelligent banana picking device is less than 8.66 mm, and the relative position error is less than 0.85%; γ The angle error is less than 1.10°, β The angle error is less than 1.25°, meeting the positioning requirements for banana picking.

Keywords: machine vision; YOLOv5; banana picking; 3-RPS parallel mechanism; banana stem recognition