Machine Learning applied to Raman spectroscopy for mineral classification on Mars. A first step toward multi-sensors analysis.
Abstract
With a Mars landing planned in February 2020 aboard NASA's next rover, Perseverance, the SuperCam instrument suite will enable remote in-situ spectral analyses [Willis et al. 2018], with three different spectroscopy techniques: Laser-Induced Breakdown Spectroscopy (LIBS), which has proved its efficiency for the analysis rock and soils as part of the Mars Science Laboratory mission [Maurice et al. 2016; Wiens et al. 2015]; Visible-IR spectroscopy; time-resolved Raman & Luminescence. These techniques present on exceptional complementarity, both on technological and scientific points of view, hence enabling the parallel acquisition of data on the chemical composition and mineralogical structure of the Martian surface.
To get the full benefit of this complementarity, we work on the combined processing of this data. We aim at taking advantage of correlations and redundancies between multi-sensors variables to improve the ranges of detection, the accuracy of classification and quantification of different elements and minerals. Previous works were performed on LIBS and Raman data fusion [Sobron et al. 2014; Rammelkamp et al. 2019]. Our work focuses on the use of decision trees and random forests models based on [Breiman et al. 1984; Breiman 2001]. We develop and test our models using a Raman database provided by the Laboratoire de Géologie de Lyon. We compare different models (using univariate or multi-variate splits, or variable selection) and discuss the geological relevance of the variables selected by the algorithms. The same algorithms are then adapted and used in a multi-sensor approach. These tools will then be applied to the study of carbonates on Mars. Carbonates attest to the presence of water in the past of Mars, and especially near Mars2020's landing site, Jezero crater [Ehlmann et al. 2009; Horgan et al. 2020], and present a great ability to preserve life forms over very long periods of time [Farmer et al. 1999], making them a very interesting lead to follow in the search for life on Mars.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMP027.0009C
- Keywords:
-
- 6225 Mars;
- PLANETARY SCIENCES: SOLAR SYSTEM OBJECTS;
- 6297 Instruments and techniques;
- PLANETARY SCIENCES: SOLAR SYSTEM OBJECTS