講演詳細

経営部門[Forest Management]

日付	2025年3月22日
開始時刻	9:15
会場名	S31
講演番号	D-23
発表題目	Tree component classification using UAV LiDAR-derived point cloud and Attention-PointNet++ deep learning Tree component classification using UAV LiDAR-derived point cloud and Attention-PointNet++ deep learning
所属	東京大学
要旨本文	Accurate classification of tree components is crucial for forest management, biodiversity monitoring, and ecological research. UAV-based LiDAR provides high-density point clouds with potential for detailed forest analysis, but segmenting these into meaningful entities remains challenging. To fill this gap, we propose a novel method using UAV LiDAR-derived point cloud data and an enhanced Attention-PointNet++ model. This approach leverages high-resolution data and attention mechanisms to improve the classification of tree components, including trunks, branches, coarse woody debris (CWD), and ground. Tested on labeled data from the University of Tokyo Hokkaido Forest (UTHF), the method achieved a high F1-score across four semantic classes, demonstrating its accuracy and efficiency. This approach has promising applications for forest inventory, monitoring, and ecosystem modeling, providing deeper insights into forest vertical structure and enhancing ecological understanding.
著者氏名	○Hu, Nan1 ・ Owari, Toshiaki2 ・ Tsuyuki, Santoshi1 ・ Hiroshima, Takuya1
著者所属	1The University of Tokyo ・ 2The University of Tokyo
キーワード	Tree component, UAV, Point cloud, Deep learning
Key word	Tree component, UAV, Point cloud, Deep learning