翟永杰, 胡东阳, 苑朝, 王家豪, 张鑫, 刘亚军. 基于视觉伺服的蝴蝶兰种苗切割系统设计与试验[J]. 农业工程学报, 2022, 38(6): 148-156. DOI: 10.11975/j.issn.1002-6819.2022.06.017
    引用本文: 翟永杰, 胡东阳, 苑朝, 王家豪, 张鑫, 刘亚军. 基于视觉伺服的蝴蝶兰种苗切割系统设计与试验[J]. 农业工程学报, 2022, 38(6): 148-156. DOI: 10.11975/j.issn.1002-6819.2022.06.017
    Zhai Yongjie, Hu Dongyang, Yuan Chao, Wang Jiahao, Zhang Xin, Liu Yajun. Design and experiments of phalaenopsis seedling cutting system using visual servo[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2022, 38(6): 148-156. DOI: 10.11975/j.issn.1002-6819.2022.06.017
    Citation: Zhai Yongjie, Hu Dongyang, Yuan Chao, Wang Jiahao, Zhang Xin, Liu Yajun. Design and experiments of phalaenopsis seedling cutting system using visual servo[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2022, 38(6): 148-156. DOI: 10.11975/j.issn.1002-6819.2022.06.017

    基于视觉伺服的蝴蝶兰种苗切割系统设计与试验

    Design and experiments of phalaenopsis seedling cutting system using visual servo

    • 摘要: 蝴蝶兰种苗自动化切割可降低组培苗染病几率,提高种苗品质。为实现蝴蝶兰种苗自动化切割,该研究针对种苗结构特性提出基于拟合直线的切点定位方法,并配合切割方法设计了弹性切割末端,搭建了基于视觉伺服的蝴蝶兰种苗切割系统。首先,采用深度学习模型对采集到的图像进行目标检测;然后,根据检测结果使用基于几何规则的切割点定位算法计算切点;最后,将切割坐标传输给切割执行机构完成切割作业。目标检测试验中,ShuffleNet v2-YOLOv5模型检测精度达96.7%,权重文件大小1.3 MB,平均检测时间0.026 s。种苗切割试验中,切割合格率高于86%,单株平均切割时间小于18 s。该系统能有效完成蝴蝶兰种苗切割任务,为蝴蝶兰组培苗自动化生产提供新思路。

       

      Abstract: Abstract: The tissue culture technology of phalaenopsis has gradually matured in recent years. Seedling stem cutting has been one of the most critical steps in the process of the tissue culture industry. An automatic cutting system can greatly contribute to reducing the probability of infection for the better quality of seedlings. In this study, a cutting-point positioning system was proposed for the automatic cutting of phalaenopsis seedlings using a visual servo. The elastic cutting end was also designed using the fitting straight line for the structural characteristics of seedlings. The seedling cutting system was mainly composed of the seedling visual detection, cutting coordinate positioning, and seedling cutting execution modules. Specifically, the seedlings were firstly photographed with an industrial camera, and then the images were transmitted and saved to the core computing unit. Shuffle Net v2 - YOLOv5 model was used for the target detection of the collected images, in order to realize the accurate detection of the seedling stem, blade, and black tuber. Secondly, the saved image was identified to evaluate the feasibility of cutting performance. The image data was then extracted from the detection frame, where the image contour in the frame was extracted to fill the maximum connected region. The stem straight-line fitting with the distance correction was used to make the connected region pixels near the black tuber position closer to the cutting center. Thirdly, the cutting-point positioning was implemented to verify the resulting data using geometric profiles. Finally, the cutting coordinate was transmitted to the cutting actuator for the cutting operation. The cutting-end design fully considered the characteristics of phalaenopsis seedlings and the cutting environment of tissue culture seedlings. The structure reduced the impact force of cutting tools when performing the cutting tasks. As such, a stable and effective operation was achieved at the cutting end for the rapid replacement of sterile consumables. A user interface was also designed to cooperate with the cutting system for the phalaenopsis seedling cutting, including the display of target detection and cutting coordinates, as well as the cutting control modules. Various components of the seedling were monitored to predict the parameters of the manipulator at each joint, particularly for the end speed over the control area. A target detection experiment was conducted to compare the Faster-RCNN, YOLOv4, YOLOv5, and MobileNet v2-YOLOv5 detection models. ShuffleNet v2-YOLOv5 model presented a higher detection accuracy of 96.7%, a weight file size 1.3 MB, and an average detection time of 0.026 s, suitable for seedling image detection. In the seedling cutting test, the cutting qualified rate was higher than 86% than before, and the average cutting time per plant was less than 18 s. The finding can lay the foundation for the phalaenopsis seedling cutting system with two-arm cooperation.

       

    /

    返回文章
    返回