Water-land classification for single-wavelength airborne LiDAR bathymetry based on waveform feature statistics and point cloud neighborhood analysis

Abstract

Accurate water-land classification in coastal zones is the basis of airborne LiDAR bathymetry (ALB)-based hydrological research and topographical map production. Considering the confusion among waveform features and the diverse geometric features of targets, it is difficult to distinguish water and land with single-wavelength ALB systems. To address these issues, a water-land classification method based on waveform feature statistics and neighborhood analysis is proposed in this paper. First, the elevations of the bimodal waveform point clouds and the thresholds calculated based on waveform feature histograms are utilized to extract coarse- and fine-scale sea surface points, respectively. Then, the thresholds, elevation variance, and geometric features in the connected region are determined to discriminate inland water points. Finally, to improve the classification accuracy, neighborhood analysis for point cloud rasterization is performed. The proposed method is verified with four swaths obtained by the Optech Aquarius ALB system near Wuzhizhou Island and Yuanzhi Island in the South China Sea. Overall accuracy values of 99.2% and 95.2% on average are obtained using the proposed method for all points and the 100 m coastline buffer, respectively. In comparison, a higher precision and shorter runtime are achieved using the proposed method than the support vector machine (SVM), random forest (RF), and fuzzy C-means (FCM) methods. Accordingly, the proposed method is a precise and efficient water-land classification method for single-wavelength ALB systems without artificial samples. In the future, this method can provide an effective technical approach for the fully automatic processing of ALB data.