LWSNet: A Point-Based Segmentation Network for Leaf-Wood Separation of Individual Trees

The accurate leaf-wood separation of individual trees from point clouds is an important yet challenging task. Many existing methods rely on manual features that are time-consuming and labor-intensive to distinguish between leaf wood points. However, due the complex interlocking structure leaves in canopy, these have not yielded satisfactory results. Therefore, this paper proposes end-to-end LWSNet separate points within canopy. First, we consider linear scattering distribution characteristics calculate local geometric with distinguishing properties enrich original cloud information. Then, fuse contextual information for feature enhancement select more representative through a rearrangement attention mechanism. Finally, use residual connection during decoding stage improve robustness model achieve efficient separation. proposed tested eight species different sizes. average F1 score as high 97.29%. results show method outperforms state-of-the-art previous studies, can accurately robustly species, sizes, structures. This study extends tree manner demonstrates deep-learning segmentation algorithm has great potential processing plant morphological traits.