Answers are grounded in this paper and cite their sources.
Impact craters are prominent lunar landforms that record impact conditions and subsequent geological modification. Semantic segmentation of their internal units, including central peaks, floors, walls, and rims, is needed for quantitative geomorphological analysis. Existing automated studies have focused mainly on crater detection or boundary extraction, whereas pixel-level mapping of crater sub-geomorphology remains limited. We developed Crater3-Net, an improved U-Net that uses a digital elevation model (DEM), slope, and plan curvature as three input channels and applies max-unpooling to recover spatial detail. On an independent test set, Crater3-Net achieved a mean intersection over union (mIoU) of 0.787 and a mean pixel accuracy (mPA) of 0.954, the highest values among the four tested configurations. We applied the trained network to lunar craters larger than 10 km and manually refined the resulting vector boundaries, producing a database of 16,412 craters and approximately 34,000 sub-geomorphological units. Within this database, the proportion of craters with mapped central peaks increased with crater diameter and broadly followed the established simple-to-complex transition. The resulting workflow supports large-scale quantitative mapping of lunar crater sub-geomorphology and provides a consistent database for constructing the Digital Moon and studying lunar surface evolution.
No takes yet. Share an insight, caveat, or question.
Deng et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: