Houyu Liang, Xiang Zhou, Tingting LV, Qingwang Liu, Zui Tao, Hongming Zhang, Meng Lei, Haili Wu
Accurate extraction of individual trees and structural attributes is crucial for assessing forest carbon stocks, ecosystem functions, biodiversity conservation, and sustainable forest management. Key individual-tree parameters, such as tree position, height, and diameter at breast height (DBH), are central to modern forest inventories. Mobile Laser Scanning (MLS) offers flexible and detailed understory measurements but remains challenged by understory occlusion, branch/leaf adhesion, and crown overlap, limiting the adaptability of existing methods. This study proposes a new MLS-based framework for individual-tree parameter extraction that integrates structural-feature-guided branch/leaf filtering and growth-oriented region growing. First, trunk–branch/leaf separation and trunk region identification were achieved using local density patterns and geometric regularity to suppress vegetation and adhesion interference. Second, trunk-based directional constraints and filtering rules derived from vertical growth characteristics guided region growing with adaptive thresholds and progressive optimization, enabling accurate segmentation of adhered trees. A reconstruction step was then applied to recover complete trunk and crown point clouds. The framework was validated on four public datasets and one self-acquired dataset (48 plots, 2,632 trees) spanning diverse forest structures. Results showed high performance, with F1-score of 0.95–0.99, Recall of 0.93–1.00, and Precision of 0.96–0.99. Height estimation achieved RMSE < 1 m ( R 2 up to 0.99) in some area, and DBH estimation achieved RMSE < 2 cm ( R 2 up to 0.95). Comparative analyses confirmed improvements over existing approaches, demonstrating enhanced generalization and interpretability. The method thus provides a reliable MLS-based solution for extracting tree location, height, and DBH in complex forest environments, supporting large-scale forest monitoring.