Real-Time Identification of Minerals from Multimodal Spectroscopy for Lunar Surface Exploration

Real-Time Identification of Minerals from Multimodal Spectroscopy for Lunar Surface Exploration. I. Drozdovskiy1, S.J. Payler1, L. Bessone1 & members of the CAVES and PANGAEA team 1Directorate of Human and Robotics Exploration, European Astronaut Centre (EAC) - European Space Agency (Linder Höhe, D-51147 Cologne, Germany; [email protected]) Introduction: Future human and robotics missions to the surface of the Moon will benefit from immediate, high-confidence recognition of geological materials to support time-sensitive scientific exploration and resource prospecting. The ESA-PANGAEA Mineralogical Toolkit addresses this challenge by developing a multimodal Machine Learning (ML)- based software framework that uniquely enables realtime in-situ mineral identification from separate and combined spectroscopic techniques [1]. The system leverages the PANGAEA Mineralogical Database (MinDB) - a comprehensive repository of reference spectra and contextual petrographic information for all known minerals identified on the Moon, Mars, and in meteorites [2]. The result is a system capable of delivering rapid state of the art mineral identification accuracy that is especially powerful when leveraging its unique ability to fuse molecular spectroscopy (Reflectance UV-to-Infrared, VNIR; and Raman) with elemental analysis (Laser-Induced-Breakdown, LIBS; and X-Rays Fluorescence, XRF spectroscopy). When synergizing with other complementary ESA projects, such as the development of the portable combined handheld spectrometers (e.g., PHOENIX — Portable Handheld cOmbinEd RamaN-LIBS-XRF [3]), communication networks, and portal computing platforms (e.g., EFB - Electronic FieldBook/ECB - Electronic Control Board [4]), the system can provide immediate decision support for important mission activities such as strategic sample triage and collection, resource deposit assessment, and quality control monitoring of in-situ resource utilization (ISRU). The project is carried out in the framework of the ESA PANGAEA course (Planetary ANalogue Geology and Astrobiology Exercise for Astronauts) [5], which beyond its educational role, also focuses on enhancing technological tools for field sampling and scientific analysis, as well as developing operational strategies for future geological field activities on the Moon and other planetary bodies. Improvements and enhancements ESAPANGAEA Mineralogical Toolkit: Over the past two years, the ESA-PANGAEA Mineralogical Toolkit has been significantly revised and improved. The updates described below enhance the ability of the system to analyze data more accurately from a greater range of sources and lays the groundwork for the system to support the analysis of the multi-modal coaligned measurements provided by the PHOENIX instrument protypes which combine Raman with LIBS/XRF spectroscopy in different hardware configurations. MinDB: Mineralogical catalog and spectral archive. The catalog of petrographic information in the MinDB has been updated to include an additional 35 minerals/endmembers which have been recently confirmed to be present in meteorites, returned samples from asteroids and the Moon, and from in-situ geochemical analyses. The catalog has also now been synchronized with recent revisions to the IMA–CNMN regarding minerals names and symbols, chemical formula, and mineral group, as well as harmonized with recently published studies on mineral surface abundance on planetary bodies and their scientific significance in context of planetary exploration (e.g., marker for important processes). Additionally, extra steps have been implemented to improve catalogue structure, reduce data redundancy and improve data integrity. The second major part of the MinDB, a customized library of reference datasets for planetary analogue minerals, has also been expanded with new reflective Visual-to-Near- & Shortwave-Infrared (VNIR), Raman, Laser-Induced-Breakdown (LIBS), and X-Rays Fluorescence (XRF) spectra. A part of these reference spectra is used to train our ML models, while another part is used for evaluating the detectability of minerals with different analytical methods. This was achieved using the most recent high-quality spectra available in open access on-line catalogues and combined with our own collection of spectroscopic measurements of planetary analogue minerals taken from different collections [2, 6]. In addition, we have established a set of internal validation mineral samples to provide a Figure 1: The current census of the minerals with archived Raman or VNIR spectra in the PANGAEA mineralogical database Moon Mars Meteorites MinDB 0 100 200 300 400 500 600 700 # Minerals # Raman # VNIR 130 101 72 101 85 86 423 253 153 651 426 268 Number of minerals with spectra Moon Mars Meteorites MinDB standardized resource for testing various prototype spectrometers, such as PHOENIX. These high-fidelity mineral samples were selected based on their ubiquity and importance for characterizing planetary geology (of the Moon, Mars, and asteroids) and supporting other exploration goals, such as in-situ resource utilization (ISRU). The selection criteria emphasized well-defined mineralogy, high purity, relevance to space analog formations, and morphological diversity. This validation set will also incorporate the rock samples specifically chosen for the ESA-DLR LUNA analog facility [7], ensuring a comprehensive testing environment. Machine Learning (ML) software for recognition of minerals from multispectral data: The major focus on improving our internally developed ML framework was not only the accuracy of the mineral recognition through multi-spectral data fusion, but also to streamline the ML operations (MLOps) workflow. For improving accuracy, we addressed a few shortcomings of our previous ML models, such as a challenge to recognize long-range spectral features via new long-range sensitive ML architectures. This is particularly beneficial for the classification of spectra covering broader spectral ranges, such as those measured by the latest the portable spectrometers, including PHOENIX [3]. We also made significant improvements to the ability of the system to combine and analyze data from different molecular (such as Raman and Reflectance) and atomic (such as XRF and LIBS) spectroscopic techniques to extract deeper insights into mineral composition and properties – a key goal of the project. Finally, we enhanced our MLOps framework to better handle joint data and accelerated code execution to speed up analysis time and reduce compute to better prepare the system to be deployed during time-sensitive activities with limited power and compute such as geological surface activities on Moon, Mars and other planetary bodies. PANGAEA Mineralogical Toolkit for Analogue Testing and Future Prospects: The integrated subsystems of the ESA-PANGAEA Mineralogical Toolkit will undergo extensive testing in field environments and at the ESA-DLR LUNA analog facility, a state-of-the-art lunar simulation environment. Integrated with other ESA funded projects, such as PHOENIX and ECB, these tests will evaluate the toolkit’s performance under realistic mission conditions, helping to ensure its readiness for deployment in future planetary exploration missions. Beyond space exploration, the toolkit also holds significant potential for terrestrial applications, such as geological research or industrial applications. 2 Dept. of Geosciences, University of Padova, Italy By combining advanced spectroscopy, machine learning, and seamless integration with complementary technologies, the ESA-PANGAEA Mineralogical Toolkit will be able to deliver rapid, accurate, and actionable insights that will empower astronauts and robotic systems alike, paving the way for more efficient and scientifically productive missions to the Moon, Mars, and beyond. Acknowledgements: We gratefully acknowledge the invaluable contributions of our collaborators: Barbaro, A.2, Joseph, F.3, Gasparetto, P.2, Kaliwoda, M.3, Nestola, F.2, Schmitt, B., Tosato, F.2, the PHOENIX team [3], and the following members of the CAVES and PANGAEA team: Adriani, I., Angellotti, A., Arbelo, P., Baranowska, M., Bardi de Fourtou, G., Brochart, A., Cook, S., Dalibot, A., Damageux, N., Delages, J., Deschand, N., Drozdovskiy, I., Eccleston, R., Fechner, R., Fechtner, K.,Franke, M., Freudenstein, A., Gade A.J., Hill, S., Huvar, O., Itzerott, M., Jahoda, P., Jaruskova, K., Kenesi, P., Laban, M., Lennert, P., Leygonie, A., Ligeza, G., Lovell, L., Massironi, M.2, Mikliss, A., Natarajan, D., Payler, S., Pozzobon, R.2, Sauro, F.2, Spedicato, S., Tatham, C., Taylor, S., Turchi, L., van Loenen, M., Vanuxem, A., Venegas, F., Vodnik, P., Zampol, I., Zhang, X., Zuncke, J.3 References: [1] Jahoda, P. et al. 2020, The Analyst, 146(1), p.184; [2] Drozdovskiy, I. et al. 2020, Data in Brief, 31, 105985; [3] Moral, A. G., et al. 2025, this conference; [4] Turchi L. et al. Planetary Space Science, 197, p.105164; [5] Sauro, F. et al. 2023, Acta Astronautica, 204, 222–238; [6] Kaliwoda, M., et al. 2025, this conference; [7] Casini, A. E. M. et al. 2020, Journal of Space Saf. Eng., 7(4), 510 3 Mineralogische Staatssammlung München, SNSB, München, Germany