Framework for the Generation of Large Datasets of Synthetic RADAR Soundings of the Martian Subsurface
Résumé
Essential for the successful training and application of AI tools to the automated processing of planetary RADAR soundings is the availability of sufficiently labeled and annotated data for the specific instruments and solar system bodies. For data of planetary RADAR sounders, no ground truth is typically available, resulting in labels and annotations derived from models of observations. A different approach is to generate synthetic data by simulation of the RADAR sounding process, including all relevant instrument characteristics and models of the surface and subsurface of the sounded solar system body. In this context, the aim of such automated processing approaches is to annotate subsurface signatures, e.g., hyperbolas, that could indicate point-like scatterers, which are of interest for a permittivity estimation. To simplify the generation of many RADAR datasets for the WISDOM instrument, a payload of the ExoMars Rosalind Franklin rover, scheduled for 2022, we have developed a framework to efficiently create random variations of models of Oxia Planums subsurface. Using this framework, we aim to generate sufficient data for the automatic processing of the WISDOM measurements and the extraction of subsurface features. Furthermore, the framework can be applied to other missions and RADAR sounders and extended to integrate synthetic results from other instruments. In this presentation, we demonstrate initial results generated with the framework, as well as introduce the general workflow and steps necessary to parametrize the model generation.