Automated surface mapping via unsupervised learning and classification of Mercury Visible–Near-Infrared reflectance spectra

Abstract

In this work we apply unsupervised learning techniques for dimensionality reduction and clustering to remote sensing hyperspectral Visible-Near Infrared (VNIR) reflectance spectra datasets of the planet Mercury obtained by the MErcury Surface, Space ENvironment, GEochemistry, and Ranging (MESSENGER) mission. This approach produces cluster maps, which group different regions of the surface based on the properties of their spectra as inferred during the learning process. While results depend on the choice of model parameters and available data, comparison to expert-generated geologic maps shows that some clusters correspond to expert-mapped classes such as smooth plains on Mercury. These automatically generated maps can serve as a starting point or comparison for traditional methods of creating geologic maps based on spectral patterns. The code and data used in this work is available as Python Jupyter Notebook on the github public repository MESSENGER-Mercury-Surface-Cassification-Unsupervised_DLR1 funded by the European Union's Horizon 2020 grant No 871149.