Laser-induced breakdown spectroscopy (LIBS) is a stand-off atomic emission spectroscopy method, which allows for quantitative and qualitative chemical analysis of materials. Since there is no need for direct sample contact, LIBS is well suited for in-situ measurements, especially in the context of planetary exploration. Currently there are three different Mars missions with active LIBS instruments on board: The ChemCam and SuperCam instrument (NASA) and the MarSCoDE instrument (CNSA).

In the past decade, the LIBS community’s interest in machine learning (ML) methods has increased steeply for several reasons:

1. Trained ML models allow for fast and automated in-situ analysis of LIBS data
2. Large LIBS data sets are available - the Curiosity rover alone has measured over 900.000 LIBS spectra over the past decade.
3. Classical analysis techniques often lack the complexity needed to account for non-linear aspects of LIBS data due to laser-surface interactions and chemical reactions in the LIBS plasma (summarized as matrix effects).

In this PhD thesis, I investigate the applicability of machine learning methods in the context of robotic in-situ exploration for chemical compound classification in LIBS spectra, generative ML models for data set extension as well as physics informed ML approaches to improve our understanding of the physical processes in the LIBS plasma.

For this purpose, I obtained a labeled LIBS data set from samples relevant for Martian exploration with our state-of-the-art LIBS setup at the DLR institute of Optical Sensor Systems in Berlin Adlershof. This data set serves as starting point for all ML models I investigate.

Finally, I wish to extend this investigation to real Martian LIBS data in order to increase the scientific return of current and future planetary exploration missions.