DeepLandforms: A Deep Learning Computer Vision toolset applied to a prime use case for mapping planetary skylights

Abstract

The exploration of terrestrial planets in the Solar System was and still is performed mainly on data that cover almost all the electromagnetic spectrum, acquired over the last century by several types of orbiters, rovers, and landers. Planetary data volumes are constantly increasing both in quality and quantity, with the contribution of both public and private entities.Imagery has always been the primary resource for researchers in planetary sciences, especially for geologists and geomorphologists. In the last two decades, the progress in the development of very high-resolution image sensors gave the community access to images with a spatial resolution on the order of centimeters. Data collected by High Resolution Imaging Science Experiment (HiRISE) instrument, on board of the Mars Reconnaissance Orbiter (MRO) (McEwen et al., 2007) or by the Narrow Angle Camera experiment, on board of Lunar Reconnaissance Orbiter (Robinson et al., 2010), have been used in several works to constrain the surface properties of respectively, Mars and the Moon.

Investigation of the Characteristic Nighttime Temperature of Potential Caves on Mars

“…A recent study reported the development of a deeplearning program that detects the cave entrance candidate automatically (Nodjoumi et al 2021). The deep learning model learns the patterns of the satellite image of cave entrance candidates and applies the patterns to the new surface regions to detect new cave entrance candidates.…”

DeepLandforms: A Deep Learning Computer Vision Toolset Applied to a Prime Use Case for Mapping Planetary Skylights

“…The original data used for this study are available at publicly data archive NASA PDS Geoscience Node ODE (PDS Geosciences Nodes, 2021). Data sets for this research are available in these in-text data citation references (Nodjoumi, 2021a) with license GPLv2+ and available at: https://doi.org/10.5281/zenodo.7351391. Software for this research is available in these in-text data citation references (Nodjoumi, 2021c): with license GPLv2+ and available at: https://doi.org/10.5281/zenodo.7488867.…”

“…Data sets for this research are available in these in-text data citation references (Nodjoumi, 2021a) with license GPLv2+ and available at: https://doi.org/10.5281/zenodo.7351391. Software for this research is available in these in-text data citation references (Nodjoumi, 2021c): with license GPLv2+ and available at: https://doi.org/10.5281/zenodo.7488867. All base images of the figures have been prepared using QGIS software 3.16 LTS and then processed with Affinity Publisher 1.1.…”

Recognition and Classification of Martian Chaos Terrains Using Imagery Machine Learning: A Global Distribution of Chaos Linked to Groundwater Circulation, Catastrophic Flooding, and Magmatism on Mars

“…OR PEER REVIEW 4 of 34 Automatic mapping and classification are a new approach in planetary geology (e.g., [25][26][27][28][29]) and can drastically reduce the effort required for investigations of surface features. Nevertheless, to date, there have been a rather limited number of examples of this approach pertaining to geological investigations of Mars (e.g., [28,[30][31][32][33][34][35]). Machine learning is increasingly used as an application in industry.…”