设施番茄采摘机器人识别定位与采摘方法研究

Recognition-Localization and Picking Methods of Facility Tomato Picking Robots

摘要:

番茄是设施农业中种植面积较大的农作物之一，在农业生产劳动力紧缺的背景下，实现其自动化采摘具有重要意义。针对设施单果番茄采摘的需求，设计了一款设施番茄智能采摘平台，主要由升降机构、采摘机构、识别与定位系统等部分组成。平台整体结构由低压一体化伺服滚珠丝杠副升降机构、六轴协作机械臂和力控末端执行器组成，实现了自动化、精确高效的采摘作业；基于改进YOLO v5-HSV融合算法来识别检测，通过对H分量进行图像阈值分割，提高了对成熟目标果实识别的准确率，有效排除未成熟番茄和枝叶背景的干扰；通过眼在手外的标定方法，使用ZED双目相机进行定位，来实时获取目标果实在机械臂基坐标系下的空间坐标。搭建的设施番茄采摘机样机平台在现场采摘试验中，识别准确率达到95.01%，采摘成功率为87.96%，单果平均采摘时间为14.56s。结果表明，YOLO v5-HSV融合算法能够减少对番茄的识别误差，使用眼在手外算法计算出转换矩阵，实现对目标果实准确识别与定位，设施番茄智能采摘装置的识别定位和采摘能力满足实际工作需求。

Abstract:

Tomatoes are one of the major crops in facility agriculture. Against the backdrop of a shortage of agricultural labor, the automated harvesting of tomatoes is of great significance. To meet the harvesting requirements of individual tomatoes in facilities, a smart harvesting platform for facility tomatoes was designed. The platform mainly consisted of a lifting mechanism, a harvesting mechanism, and an identification and positioning system. The overall harvesting structure was composed of a low-voltage integrated servo ball screw pair lifting mechanism, a six-axis collaborative robotic arm, and a force-controlled end effector, achieving automated, precise, and efficient harvesting operations. Based on an improved YOLO v5-HSV fusion algorithm for identification and detection, image threshold segmentation was performed on the H component. This improved the accuracy of identifying ripe target fruits and effectively eliminated the interference of unripe tomatoes and leafy backgrounds. Using the eye-in-hand calibration method, the ZED stereo camera was employed for positioning to obtain the spatial coordinates of target fruits in the robotic arm’s base coordinate system in real-time. In field harvesting experiments, the facility tomato harvesting robot prototype built achieved a recognition accuracy of 95.01%, increased the harvesting success rate to 87.96%, and reduced the average harvesting time per fruit to 14.56s. The results showed that the YOLO v5-HSV fusion algorithm can reduce recognition errors of tomatoes. The eye-in-hand algorithm was used to calculate the transformation matrix for accurate identification and positioning of target fruits. The recognition, positioning, and harvesting capabilities of the smart harvesting device for facility tomatoes met the practical operational requirements.