Deep Learning Analysis of UAV Lidar Point Cloud for Individual Tree Detecting

In this paper, we address the important task of Individual Tree Detection (ITD) in forest environments for enabling tree parameter estimation including tree count, height, volume, and crown dimensions. Recent advances in high-resolution multispectral and LiDAR data collected by Unmanned Aerial Vehicles (UAVs) show a promising solution for ITD. We introduce a novel ITD method using the YOLO V7-tiny deep learning framework on UAV LiDAR data. First, we rasterize point clouds into Vertical Density (VD) and Canopy Height Models (CHM), and then we utilize the modified YOLO V7tiny algorithm to detect the boundary of the trees. The accuracy, precision, recall, and F1-score results of YOLO V7 compared to YOLO3 showed a significant improvement. The proposed method demonstrates promising results for urban and forest tree inventory updates and contributes to largescale satellite-based forest structure and biomass estimation.