Authors: Cheng-Zhi Qin*, Institute of Geographic Sciences & Natural Resources Research, CAS, Yan-Wen WANG, State Key Lab of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, CAS, Wei-Ming Cheng, State Key Lab of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, CAS, A-Xing Zhu, University of Wisconsin-Madison
Topics: Geographic Information Science and Systems, Geomorphology, Quantitative Methods
Keywords: Automatic crater detection, random forest, DEM, spatial structure
Session Type: Paper
Start / End Time: 3:05 PM / 4:45 PM
Room: Roosevelt 3, Marriott, Exhibition Level
Presentation File: No File Uploaded
Automatic crater detection is important for planetary geology and space exploration. Crater detection approaches (CDAs) based on digital elevation model (DEM) could consider abundant terrain information depicting both shape and spatial structure of craters, thus have advantages over those CDAs based on remote sensing photographs. However, existing CDAs based on DEM are mainly designed to detect the depressions with round shape or simplified spatial structure of conceptual crater, and still cannot effectively consider the spatial structural information of craters in reality. In this abstract we propose a novel automatic CDA which is based on random forest classifiers trained with existing crater map and spatial structural information derived from DEM. In the proposed approach, existing crater map of training area is used to provide samples for training random forest classifiers, when corresponding input features are spatial structural information derived from DEM. By such way, the spatial structural information of craters in reality (especially those within real craters) can be effectively considered by the proposed CDA. A case study was conducted with a lunar catalog compiled by experts based on Lunar Orbiter Laser Altimeter data and a Chang’E-1 lunar DEM with 500 m resolution. Experimental result shows that the proposed approach performed better than a representative of existing CDAs based on DEM (i.e., the AutoCrat approach).