Abstract :
[en] In recent times, there has been a resurgence of interest in not only revisiting the
Moon but also establishing a lasting and sustainable human presence there. Numer-
ous agencies and private corporations, gearing up for future lunar missions, have
established key goals involve both scientific exploration and the emerging lunar
economy. These objectives include carrying out scientific experiments, harvesting
and utilizing lunar resources on-site, and examining potential methods of using the
Moon as an efficient launchpad to go further into the Solar System. In facilitating
these lunar missions, robotics emerges as the main disruptive technology. It can
allow the execution of various mission-critical activities in harsh environments,
ensuring mission success without jeopardizing human safety.
This thesis addresses the challenge presented by the limited resolution of lu-
nar data and the overwhelming amount of non-processed information provided
by the different remote sensing missions. Robotic operations on the Moon require
tremendous precision in the mission planning phase. For this purpose, the remote
sensing data collected by various satellites must be processed and relayed to the
mission planning teams in the highest possible resolution and in an easily under-
standable and digested format. To address this issue, the research presented on this
thesis adopts a Machine Learning (ML) approach to enhance and process lunar data
gathered by different satellite sensors, providing more detailed and comprehensive
insights into the lunar surface, which is crucial for future robotic mission planning.
ML provides the necessary tools to efficiently handle and analyze vast volumes of
data, a critical aspect in deriving meaningful results, reaching valid conclusions,
and deepening our understanding of the subject under study. In this thesis, the
author suggests two distinct methods for enhancing and increasing the resolution
of lunar images to facilitate improved robot navigation on the lunar surface. The
first method involves creating a training dataset using a digital analog environment
and utilizing multiple frames of the same location for image enhancement. The
second approach proposes a unique architecture that utilizes a single capture
from the lunar surface for resolution upscaling, accompanied by an uncertainty
estimation of the process. Lastly, a thermophysical analysis of the lunar surface is
conducted, which involves processing lunar thermal data in the search for recent
asteroid impacts on the lunar surface.