Planetary Mineral Database Development and Validation of Spectra Classification Methods

Sammanfattning: Future astronauts working on planetary bodies need to have a detailed understanding of their surroundings and get real-time feedback from their tools to properly explore these bodies. As such, the ESA PANGAEA course trains astronauts in astrobiology and geology to help with their research as well as providing analytical techniques for sampling and scientific analysis in the field. A novel Machine Learning (ML) software has been developed in collaboration with the ESA/EAC CAVES & PANGAEA team to build a suitable spectral classification and identification method for planetary materials. Since the beginning of this project in 2020, the development of the software has significantly increased in its capability and the framework for testing the machine learning model more complex. This internship focuses on developing and enhancing the machine learning software for classifying minerals using the visual to near infrared spectra from ESA’s mineralogical database. The developed model was utilised during the PAGNAEA 2021 astronaut geological field training and was successful at classifying 16 minerals. In addition, two novel classification methods, XGBoost and Transfer Learning, have been implemented and thoroughly tested. The latter technique was integrated utilising both online models as well as creating and transferring our own models. Transferring our own model increased the mineral classification accuracy in smaller mineral subsets. However, this initial increase in classification accuracy is substantial as there are many avenues to optimise the method. I am excited to see how future members of the CAVES & PANGAEA team develop our software and will be available for communication for the foreseeable future.