Using Synthetic Tree Data in Deep Learning-Based Tree Segmentation Using LiDAR Point Clouds

Deep learning, neural networks and other data-driven processing techniques are increasingly used in the analysis of LiDAR point cloud data forest environments due to benefits offered accuracy adaptability new environments. One downsides these practical applications is requirement for manually annotated necessary training networks, which can be time consuming costly attain. We develop an approach tree stem segmentation from clouds that uses synthetic a custom simulator, generate large quantities examples without manual human effort. Our simulator captures geometric characteristics stems foliage, automatically-labelled generated semantic algorithm based on PointNet++ architecture. Using evaluations real aerial terrestrial range different sites, we demonstrate our data-trained models out-perform, or provide comparable performance with trained sites when available limited (increases IoU 1–7%). simulation code open-source made research community.