Multi-information PointNet++ fusion method for DEM construction from airborne LiDAR data

Abstract

Airborne light detection and ranging (LiDAR) is a popular technology in remote sensing that can significantly improve the efficiency of digital elevation model (DEM) construction. However, it is challenging to identify the real terrain features in complex areas using LiDAR data. To solve this problem, this work proposes a multi-information fusion method based on PointNet++ to improve the accuracy of DEM construction. The RGB data and normalized coordinate information of the point cloud was added to increase the number of channels on the input side of the PointNet++ neural network, which can improve the accuracy of the classification during feature extraction. Low and high density point clouds obtained from the International Society for Photogrammetry and Remote Sensing (ISPRS) and the United States Geological Survey (USGS) were used to test this proposed method. The results suggest that the proposed method improves the Kappa coefficient by 8.81% compared to PointNet++. The type I error was reduced by 2.13%, the type II error was reduced by 8.29%, and the total error was reduced by 2.52% compared to the conventional algorithm. Therefore, it is possible to conclude that the proposed method can obtain DEMs with higher accuracy.