Airborne Lidar Point Cloud Classification fusing Spectral Information

Abstract

Airborne lidar scanning technology can quickly obtain a large amount of three-dimensional coordinate information on the surface of ground objects. However, due to the disorder and sparseness of point cloud, how to efficiently process the point cloud has become a research hotspot. In order to achieve a more accurate point cloud classification and solve the problem that the inefficient classification is difficult to meet the follow-up processing requirements of point cloud caused by the lack of point cloud information, an airborne lidar point cloud classification method combining spectral information is proposed. Pointnet ++ is used as the basis of the network. As the perspective changes, we enlarged the radius of the extracted sphere neighborhood and improved the segmentation range of the network input subset. In order to improve the distinction of points, three-dimensional information, laser intensity information and spectral information were fused to make the fused data set. The results of the experiment using the Vaihingen regional benchmark airborne LiDAR point cloud sets provided by the International Society for Photogrammetry and Remote Sensing (ISPRS) show that the overall classification accuracy reached 86.21% after the fusion of spectral information, which was 10.02% higher than that before the fusion. The fusion of spectral information can effectively improve the classification effect and provide reliable information for the follow-up processing of airborne lidar point cloud.