A peduncle detection method of tomato for autonomous harvesting

Abstract For automating the harvesting of bunches tomatoes in a greenhouse, end-effector needs to reach exact cutting point and adaptively adjust pose peduncles. In this paper, method is proposed for peduncle localization estimation. Images captured real time at fixed long-distance are detected using YOLOv4-Tiny detector with precision 92.7% detection speed 0.0091 s per frame, then YOLACT + Network mAP 73.1 0.109 frame used segment close-up distance. The segmented mask fitted curve least squares three key points on found. Finally, geometric model established estimate an average error 4.98° yaw angle 4.75° pitch over 30 sets tests.