Detecting Craters in Images Using Machine Learning and Classical Techniques for Lunar Navigation

Loss of ground-based communication is an increasing concern for crewed and uncrewed spacecraft as cislunar traffic grows. To address navigation challenges in lunar environments, a Target & Range-adaptive Optical Navigation system (TRON) was developed to enable onboard navigation independent of ground communication. One component of the TRON suite is a technique designed specifically for terrain relative navigation (TRN) through the identification of lunar craters from orbital altitudes. The effectiveness of this method is directly tied to crater detection performance, as many false positives or false negatives can deteriorate the navigation results downstream. This paper presents a combination of machine learning and traditional computer vision methods for crater identification with built-in fault prevention. The approach is evaluated on synthetic imagery from Gateway's NRHO orbit and real lunar imagery collected during the Artemis I mission, showing promising results and establishing a clear path toward future flight demonstrations.