Identification of the operating position and orientation of a robotic kiwifruit pollinator

Abstract

Accurate identification of flower position and orientation is an essential prerequisite for robotic pollinators to accomplish precise targeted pollination and improve the pollen utilisation rate. To accurately identify the pollination position and orientation of kiwifruit flowers, this study presents a vision-based information perception approach for automatically detecting flowers and accurately identifying their positions and orientations. After using YOLO v4 to identify the kiwifruit flower, the contour detection algorithm was used to detect the contour boundary of the flower pistil area and the contour boundary of the minimum circumscribed circle of the petals, and fit the centres of gravity of the pistil contour and the petal contour according to the two contours respectively. The line connecting the two centre points was used as the central axis representing the growth direction of the flowers, achieving precise identification of the flower position and orientation. After building a robotic pollinator, an experiment was conducted to identify kiwifruit flower targets and a performance test experiment was performed to target pollinated kiwifruit flowers in a kiwifruit orchard with a mechanical arm based on flower position and orientation. The experimental results showed that the precision of the identification of kiwifruit flowers was 95.27%, the success rate of the mechanical arm for target pollination based on flower position and orientation identification was 89.59%, the average pollination time was 6 s⋅per flower, and the pollination efficiency was 20 h⋅ha-1. The identification algorithm of flower position and orientation proposed in this study can improve the precision and success rate of robotic pollination.