DEEP LEARNING UNIVERSAL CRATER DETECTION USING SEGMENT ANYTHING MODEL (SAM)

Craters are amongst the most important morphological features in planetary exploration. Detecting, mapping and counting craters is traditionally performed manually, which is laborious and time-consuming. Existing machine learning (ML) approaches for automated crater detection are typically trained on specific data types (e.g., DEMs, orbiter images and metadata), limiting their reliability and applicability across different sources, angles, and setups. This paper presents a universal crater detection scheme based on META AI's Segment Anything Model (SAM), a prompt-able segmentation system with zero-shot generalization to unfamiliar objects and images without additional training. Using SAM, crater-like objects are identified across diverse data types (e.g., raw satellite imagery, Level-1/2 products, DEMs) and planetary setups (e.g., Moon, Mars) and capture geometries. Geometrical shape indexes retain only crater-like masks, which are then fitted with ellipses to recover crater location and size/geometry. Case studies on Lunar, Martian, and Phobos datasets demonstrate the effectiveness and universality of the approach.

Iraklis Giannakis, Anshuman Bhardwaj, Lydia Sam, Georgios Leontidis