Predicting Vegetation Stratum Occupancy from Airborne LiDAR Data with Deep Learning

We propose a new deep learning-based method for estimating the occupancy of vegetation strata from airborne 3D LiDAR point clouds. Our model predicts rasterized maps three corresponding to lower, medium, and higher cover. weakly-supervised training scheme allows our network only be supervised with values aggregated over cylindrical plots containing thousands points which are typically easier produce than pixel-wise or point-wise annotations. employ neural operating on points, whose prediction projected onto rasters representing different strata. outperforms handcrafted, regression learning baselines in terms precision by up 30%, while simultaneously providing visual interpretable predictions. provide an open-source implementation along dataset 199 agricultural train evaluate weakly algorithms.