An apple tree pruning robot system based on branch segmentation and decision-making control

Intelligent pruning is an important part of smart orchard management operations, and the study of intelligent pruning robot is of great significance for the construction of smart orchards. This paper proposes an intelligent fruit tree pruning robot system for tall spindle apple trees, which consists of a vision system, a pruning decision-making system, and a robotic arm with an end-effector control system. The vision system is employed for apple tree branch segmentation, utilizing the SOLOv2 Swin algorithm, which achieves a maximum mAP50 segmentation accuracy of 0.941. Compared to other segmentation algorithms, this method significantly improves segmentation accuracy and performance. The pruning decision-making system determines target branches for pruning by analyzing the diameter and spacing of the branches, based on pruning rules, and simultaneously determining the pruning point. The robotic arm with end-effector control system implemented path planning using the Rapidly exploring Random Tree (RRT) algorithm improved by a dynamic step-size strategy and B-spline curve (DB-RRT). The results demonstrated that this algorithm reduced the path length and planning time by 5.5% and 33% respectively compared to RRT algorithms, and achieving optimal path smoothness. The system was tested in both lab and field settings. Lab experiments achieved a pruning point detection rate of 86.7% and a pre-pruning success rate of 80.0%. In the field experiment, 11 of the 15 target branches were successfully pruned, achieving an overall success rate of 73.3%. Among these, five branches were pruned on the first attempt, five on the second attempt, and one on the third attempt. The average pruning time of 17.8 s per branch. The results indicate that the system can effectively prune spatial fruit trees, providing a technical foundation for intelligent fruit tree pruning.