A New Strategy for Individual Tree Detection and Segmentation from Leaf-on and Leaf-off UAV-LiDAR Point Clouds Based on Automatic Detection of Seed Points

Accurate and efficient estimation of forest volume or biomass is critical for carbon cycles, management, the timber industry. Individual tree detection segmentation (ITDS) first key step to ensure accurate extraction detailed structure parameters from LiDAR (light ranging). However, ITDS still a challenge achieve using UAV-LiDAR (LiDAR Unmanned Aerial Vehicles) in broadleaved forests due irregular overlapped canopies. We developed an framework based on point clouds. It involves ITD (individual detection) clouds taken during leaf-off season, initial ITS segmentation) seed points ITD, improvement through refining process. The results indicate that this new proposed strategy efficiently provides ITDS. show following: (1) point-cloud-based methods, especially Mean Shift, perform better selection than CHM-based (Canopy Height Model) methods seasons; (2) significantly improved accuracy efficiency algorithms; (3) process DBSCAN (density-based spatial clustering applications with noise) kNN (k-Nearest Neighbor classifier) classification reduced edge errors results. Our study novel demonstrates proficiency dense deciduous forests, could be applied single-phase instead multi-temporal data future if have trunk points.