3DEPPCC: An automated DL-based point cloud classification tool for 3DEP point clouds
This toolkit will enhance 3DEP classification accuracy and automation, broadening its usability to external users
The U.S. Geological Survey’s (USGS) 3D Elevation Program (3DEP) is collecting lidar point data for the entire continental U.S., One of the attributes that make the data useful is classification of the points as ground, vegetation, buildings, and other landscape categories. Current minimum classification for 3DEP is identifying points as ground, non-ground, and unclassified. Accurately classified point clouds can support many down-stream applications such as in hydrologic analysis, urban planning, and forest management. The current 3DEP classification specifications result in data that can be useful for Digital Terrain Model (DTM) extraction; however, more detailed landscape segmentation is not viable. Deep Learning (DL)-based approaches have been developed for point classification with outstanding performance. This project will employ proven DL technologies in the development of a user-friendly open-source toolkit that will automate point cloud classification to refine and enrich the attributes of existing and future 3DEP data.
This toolkit will enhance 3DEP classification accuracy and automation, broadening its usability to external users
The U.S. Geological Survey’s (USGS) 3D Elevation Program (3DEP) is collecting lidar point data for the entire continental U.S., One of the attributes that make the data useful is classification of the points as ground, vegetation, buildings, and other landscape categories. Current minimum classification for 3DEP is identifying points as ground, non-ground, and unclassified. Accurately classified point clouds can support many down-stream applications such as in hydrologic analysis, urban planning, and forest management. The current 3DEP classification specifications result in data that can be useful for Digital Terrain Model (DTM) extraction; however, more detailed landscape segmentation is not viable. Deep Learning (DL)-based approaches have been developed for point classification with outstanding performance. This project will employ proven DL technologies in the development of a user-friendly open-source toolkit that will automate point cloud classification to refine and enrich the attributes of existing and future 3DEP data.