A free and open-source tool for the automatic detection of mineral phases through the clustering of NIR hyperspectral images on Mars

Authors:
Preprint
Publisher website
Google Scholar
Abstract
Spectral and hyperspectral data provide essential information on the surface composition of rocky planetary bodies, through which we can reconstruct and understand their geologic evolution and environmental history. The study of compositional data is commonly approached through the definition and mapping of spectral parameters such as band depths, band areas, and spectral slopes. Applications of more advanced tools, such as machine learning-based automatic detection systems are steadily increasing in the planetary science field. We present a newly developed, Python-based, open source tool to perform unsupervised clustering on CRISM near-infrared (NIR) hyperspectral data. The tool employs two different algorithms: k-Means and Gaussian Mixture Models. Data can be fed to these algorithms following some pre-processing steps and dimensionality reduction, which include the use of projection-based and manifold-based dimensionality reduction techniques (Principal Component Analysis and Uniform Manifold Approximation and Projection). With this tool, we identify and cluster a great variety of spectral signatures, ranging from mafic minerals with broad absorption bands to Fe/Mg phyllosilicates and sulfates, which exhibit multiple narrow absorption features. We showcase the abilities of the tool on a set of selected CRISM images. Results highlight how automated clustering methods can greatly support and enhance the findings derived from classic analysis of spectral parameters. Moreover, with minor tuning and modifications, the tool can be broadly applied both to different spectral ranges and to spectral or hyperspectral data from any planetary surface.