查看文章

Introduction: NASA’s Curiosity rover has been traversing Gale crater, a lacustrine region chosen as the landing site of the rover due to its potential for past habitability, since landing there in 2012 [1]. Curiosity spent the first~ 760 martian solar days (sols) of the mission in the Bradbury formation, an ancient fluvio-lacustrine system [1, 2].

In the nine years since landing on Mars, Curiosity has observed a wide variety of rock types, and several classification methods have been developed with the aim of sorting these rocks into process-oriented facies [eg, 3-5]. But accurate classification of rocks can be challenging when information is limited to images and chemical composition, meaning process-oriented classifications risk introducing bias.[6] addressed this issue by developing a classification system based only on simple visual attributes. This system involved three phases: first,(1) the manual process of reviewing each target’s RMI and encoding its visual attributes as a 17-digit binary number; then (2) an initial algorithmic grouping of the targets; and finally (3) a manual review, which refined the algorithmgenerated groupings [6]. The second phase generated ten graph components with varying connectivity, and the relatively weakly connected components seemed to correlate with worse target image association [6]. In this work, we attempt to automate the third phase by automatically partitioning the components with weak connectivity. Methods: Curiosity’s ChemCam instrument uses Laser-Induced Breakdown Spectroscopy (LIBS) to obtain chemical information about rock targets [7, 8]. With each LIBS analysis, high-resolution images of the target are …