A deep neural network based on U-Net for classifying airborne LiDAR data

LiDAR point cloud classification has always been a difficult task. Deep learning (DL) models have recently been widely used in computer vision studies, thanks to rapid advances in machine learning technology. In recent years, deep neural networks (DNN) have been used for classification and segmentation of LiDAR point clouds. We proposed a Fully Convolutional Network (FCN) architecture that can classify LiDAR data. This workflow is built around converting a three-dimensional (3D) LiDAR point cloud to a single two-dimensional (2D) multi-channel image. We tested the performance of our neural network model on the DALES airborne LiDAR dataset. The results showed that our network model performed well when compared to the LAStools software, with an average F1 score of 86% for our best model across all three classes.