Leaf and Wood Classification in Southern Pines Trees Using High Resolution Terrestrial Laser Scanning Data

Forest attributes at the tree-level could be quickly and precisely acquired using terrestrial laser scanning (TLS). For a quantitative analysis of forest attributes, including timber volume and leaf area index, the leaf and wood part of the high resolution TLS-derived 3D point cloud should be separated. There are several strategies used for classifying these two components, but no studies so far have tested them in southern pine forest in the U.S. Herein, one longleaf pine tree was used to test three geometric feature-based (TLSeparation, lidUrb-graph and lidUrb-dbscan) wood/leaf classification algorithms from the TLS data. TLSeparation's overall accuracy is 74.65%, lidUrb-graph is 82.74%, and lidUrb-dbscan is 83.03%. The performance of a segment-wise approach (e.g. lidUrbAr-dbscan) is superior to the point-wise algorithm (e.g. TLSeparation and lidUrb-graph), but none of them achieves the same high precision as the results with broadleaf trees. Southern pine trees have not been well studied or treated using current approaches.