This study presents an advanced hybrid LiDAR-based framework for individual tree segmentation and classification in complex forest environments, addressing key challenges related to high-density point cloud processing and structural variability of vegetation. The proposed approach integrates multi-return LiDAR point cloud analysis with canopy height modeling (CHM), enabling accurate representation of vertical forest structure and improved discrimination between canopy layers. A multi-stage processing pipeline is implemented, including noise filtering, ground classification, and the derivation of digital terrain and surface models to support precise canopy height estimation. For tree delineation, a combination of clustering-based segmentation techniques and watershed algorithms is employed to identify individual tree crowns in both homogeneous and heterogeneous forest stands. The extracted tree-level features—such as height, crown width, point density, and intensity distribution—are further utilized within machine learning classification models, including ensemble methods and deep learning architectures, to enable species-level discrimination. Experimental results demonstrate high segmentation accuracy in dense forest conditions, where traditional field-based and image-based methods often fail due to occlusion and limited spatial resolution. The proposed framework significantly improves scalability and computational efficiency, making it suitable for large-area forest monitoring applications. Furthermore, the integration of LiDAR-derived structural metrics enhances the reliability of forest inventory processes and supports advanced applications in biomass estimation, ecosystem monitoring, and sustainable forest management.

LIDAR,Tree segmentation, point cloud, clustering, machine learning

environmental engineering

Presentation: oral presentation