An adaptive method for individual tree segmentation synthesizing canopy cover and competitive mechanism using UAV data

Abstract

Accurate individual tree segmentation (ITS) is crucial for precision forestry and small-scale carbon sink accounting; however, canopy overlap in complex forest stands—particularly in northern plantations, presents substantial challenges for conducting ITS using LiDAR point cloud. This study introduces an adaptive ITS method that incorporates canopy cover as the primary constraint in marker-controlled watershed segmentation. This addresses two typical segmentation biases: low canopy cover areas that are prone to under-segmentation are refined using the DBSCAN spatial clustering to recover missed tree boundaries, whereas high canopy cover regions that were prone to over-segmentation were optimized using Hegyi index-enhanced improved K-means clustering method of raw point cloud data for context-aware region merging. By fusing the canopy height model (CHM) efficiency for rapid canopy contour extraction with point cloud-derived 3D structural details, this “cover-degree-driven, scene-adaptive” method balances computational speed and segmentation precision. The method was validated across 28 plots, the method achieving F1 scores of 0.89–0.95 for four tree species and outperforming traditional ITS methods in mixed forests with F1 improvements of 0.12–0.24. This method enhances the ITS accuracy of individual tree aboveground biomass estimation, thereby directly facilitating efficient small-scale carbon accounting, streamlined forest inventories, and sustainable precision management in complex ecosystems.