Accurate Forest Mapping Using TreeLearn and Lidar-Based Point Clouds

This study introduces TreeLearn, a 3D deep learning framework for segmenting and classifying individual trees in Lidar-based forest point clouds. Using high-resolution UAV, TLS, and MLS datasets from diverse global forests, TreeLearn distinguishes between tree and non-tree points with high accuracy. It applies offset prediction and clustering to identify tree instances and was tested across different forest types to evaluate domain transferability. Results show that TreeLearn effectively handles complex canopy structures and variable data sources, setting a new benchmark for automated forest mapping and analysis.