Individual Tree Segmentation Using Deep Learning and Climbing Algorithm: A Method for Achieving High-precision Single-tree Segmentation in High-density Forests under Complex Environments

Accurate individual tree segmentation, which is important for forestry investigation, is still a difficult and challenging task. In this study, we developed a climbing algorithm and combined it with a deep learning model to extract forests and achieve individual tree segmentation using lidar point clouds. We tested the algorithm on mixed forests within complex environments scanned by unmanned aircraft system lidar in ecological restoration mining areas along the Yangtze River of China. Quantitative assessments of the segmentation results showed that the forest extraction achieved a kappa coefficient of 0.88, and the individual tree segmentation results achieved F-scores ranging from 0.86 to 1. The climbing algorithm successfully reduced false positives and false negatives with the increased crown overlapping and outperformed the widely used top-down region-growing point cloud segmentation method. The results indicate that the climbing algorithm proposed in this study will help solve the overlapped crown problem of tree segmentation under complex environments.