Machine cataloging of impact craters on Mars

Abstract

This study presents an automated system for cataloging impact craters using the MOLA 128 pixels/degree digital elevation model of Mars. Craters are detected by a two-step algorithm that first identifies round and symmetric topographic depressions as crater candidates and then selects craters using a machine-learning technique. The system is robust with respect to surface types; craters are identified with similar accuracy from all different types of martian surfaces without adjusting input parameters. By using a large training set in its final selection step, the system produces virtually no false detections. Finally, the system provides a seamless integration of crater detection with its characterization. Of particular interest is the ability of our algorithm to calculate crater depths. The system is described and its application is demonstrated on eight large sites representing all major types of martian surfaces. An evaluation of its performance and prospects for its utilization for global surveys are given by means of detailed comparison of obtained results to the manually-derived Catalog of Large Martian Impact Craters. We use the results from the test sites to construct local depth–diameter relationships based on a large number of craters. In general, obtained relationships are in agreement with what was inferred on the basis of manual measurements. However, we have found that, in Terra Cimmeria, the depth/diameter ratio has an abrupt decrease at ~38°S regardless of crater size. If shallowing of craters is attributed to presence of sub-surface ice, a sudden change in its spatial distribution is suggested by our findings.