Automatic Machine Learning Classification Applied to Dawn/VIR Data in View of MERTIS/BepiColombo

Abstract

Remote sensing spectroscopy is one of the most commonly used technique in planetary science and for recent instruments producing huge amount of data, classic methods could fails to unlock the full scientific potential buried in these measurements. We explored several Machine Learning techniques: A multistep clustering method is developed, using an image segmentation method, a stream algorithm, and hierarchical clustering. The MErcury Radiometer and Thermal infrared Imaging Spectrometer (MERTIS) is part of the payload of the Mercury Planetary Orbiter spacecraft of the ESAJAXA BepiColombo mission. MERTIS’s scientific goals are to spectrally identify rockforming minerals, to map the surface composition, and to study surface temperature variations on Mercury. To cope with the stream of data that will be delivered by MERTIS, we developed an algorithm that could aggregate new data as they are acquired during the mission. This give the scientist a guide for the most interesting features on Mercury without being lost in highvolume dataset. The NASA mission DAWN carries a suites of instruments aimed at understanding the two most massive objects in the main asteroid belt: Vesta and Ceres. DAWN has already successfully completed the exploration of Vesta in September 2012 and it is now in the extended mission phase around Ceres. The DAWN/VESTA VIR data are a testbed for the algorithm developed for MERTIS. The algorithm identified the olivine outcrops around two craters on Vesta’s surface described in. We furthermore mimic the data acquisition process as if the mission were dumping the data live with a data stream cluster algorithm, analyzing one datacube and sequentially add the remaining data. The algorithm provides insightful information on the novelty and classes in the data as they are collected. This will enhance MERTIS targeting and maximize its scientific return during BepiColombo mission at Mercury.