An Autonomous UAV System for Precision Spraying of Seedling Pines to Prevent Moose Damage

We propose a novel and comprehensive autonomous unmanned aerial vehicle (UAV) system for precision and targeted spraying of seedling pine trees. The system includes three main modules: real-time state estimation, map optimization with individual tree detection, and path planning. All computations are performed onboard the UAV using a single-board computer. An IMU, a LiDAR, and optionally a barometer are the only sensors used in the proposed system. A pose-graph approach fuses data from these sensors in real time to estimate the vehicle’s state. Keyframes are selected from LiDAR scans, and individual trees are segmented using a region-growing algorithm. The keyframes are also used to generate loop-closure candidates, which help reduce long-term drift in state estimation. A path-planning framework uses detected seedling trees to plan a path from the top of one tree to the top of another, while avoiding surrounding obstacles. The proposed framework is tested in real-world forest environments, demonstrating accurate tree segmentation, coherent mapping, precise state estimation, and autonomous navigation for spraying.