Seabed Segmentation of Airborne Bathymetric Light Detection and Ranging Point Cloud Using Window-based Attention and Orthogonal Regularized PointNet

Seabed segmentation from airborne bathymetric Light Detection and Ranging (LiDAR) point cloud data presents unique challenges, primarily due to variations in the z-axis resulting from differences in water depth and seabed topography. To address these complexities, we introduced an improved version of PointNet specifically designed for seabed segmentation using Airborne Bathymetric LiDAR (ABL) point cloud data. The proposed method integrates a window-based attention mechanism to capture spatial relationships in both horizontal and vertical dimensions while incorporating orthogonal regularization to preserve geometric integrity. The model's performance was assessed using various normalization methods and window sizes, demonstrating its effectiveness in accurately identifying seabed regions. Experimental results indicate that while the proposed network generally improves segmentation accuracy, its performance is sensitive to the choice of normalization and window parameters. This study represents a meaningful advancement in applying deep learning techniques to bathymetric LiDAR data, offering a robust framework for seabed segmentation.