Imaging Mars analog minerals' reflectance spectra and testing mineral detection algorithms

Mineral mapping of a planetary surface is of prime importance to better understanding its composition and constraining its geologic evolution. Visible-to-Near-Infrared (VNIR) hyperspectral remote sensing is the primary tool to detect the composition of the martian surface at high spatial resolution. However, it is not straightforward to validate the mapping results since mineralogical groundtruth is typically unavailable. Furthermore, determining whether common approaches used to decode terrestrial VNIR data are applicable to Mars requires rigorous evaluation. In this work, we collected a suite of laboratory hyperspectral images with groundtruth, consisting of binary mixtures of Mars Global Simulant (MGS-1) and various hydrous minerals. The spectral effects of mixing MGS-1 with hydrous minerals were analyzed through conventional spectral parameters. It is observed that MGS-1 substantially reduces the spectral contrast of hydrous minerals. We then conducted a systematic comparison of seven benchmark target detection algorithms on the laboratory datasets, including Constrained Energy Minimization, Adaptive Coherence Estimator, Matched filter, Hierarchical Constrained Energy Minimization, Sparse Representation for Target Detection, Joint Sparse Representation and Multi-task Learning method, and Multi-task Joint Sparse and low-rank Representation (MTJSLR) method. Extensive experiments demonstrate that MTJSLR outperforms other competitors in successfully detecting low modal abundance hydrous minerals. The detection limits for nontronite, montmorillonite, kaolinite, gypsum, and calcite are 2.5%, 2.5%, 1%, 5%, and 5%, respectively. Carbonates are more challenging to detect than phyllosilicates even at the same abundance, which could partially contribute to the relative paucity of observed carbonates on Mars relative to phyllosilicates.

Automated summary

Mineral mapping of a planetary surface is crucial for understanding its composition and evolutionary history. Hyperspectral remote sensing is a primary tool for detecting martian surface composition, but validating mapping results is challenging due to lack of groundtruth. This study compared target detection algorithms on laboratory datasets and found that MTJSLR outperforms others in detecting low abundance hydrous minerals, with carbonates being more difficult to detect than phyllosilicates on Mars.