With the aim of enhancing autonomous orbit determination (OD) capabilities of lunar navigation satellites, this paper proposes a visual processing technique called Crater Matching, based on machine learning. Specifically, the paper analyzes the performance results obtained from different architectural implementations of convolutional neural networks (CCNs) and the hardware used to identify an on-board feasible solution. The research focuses on achieving real-time processing and on-board execution of the crater matching algorithm, ultimately enhancing the OD autonomy of navigation satellites in lunar environments