Towards Automated Mineral Identification in Martian rocks from X-ray Diffraction Patterns
Abstract
The CheMin (Chemistry and Mineralogy) instrument on the Curiosity rover has provided a rich set of X-ray diffraction (XRD) patterns from Martian rocks and regolith. These XRD patterns have allowed geologists to make exciting new discoveries about the mineralogy and the geological history of Mars. These discoveries pave the way for further Martian exploration and provide a deeper understanding of Martian geology. The Curiosity rover is very slow by design, travelling at about 4 cm/s. New, faster rovers are being developed to increase scientific throughput and exploration. XRD is valuable for future missions as it can produce new discov- eries and be used to find interesting locations for further investigation. However, the length of time it takes for XRD data to be sent to Earth and analyzed by an expert is a bottleneck that lim- its the usefulness for these types of instruments. Instead, automatic analysis of XRD patterns onboard the rover could be used to identify minerals and inform navigation when looking for scientifically interesting materials. To this end, an XRD dataset of 42,000 XRD patterns with up to 15 out of 84 possible minerals in each pattern was created using extensive physics-based augmentation. Several convolutional neural networks (CNN) with different architectures were trained and compared on this dataset. The best model was then tested on 45 CheMin XRD pat- terns and 7 Martian analogue patterns, obtaining an F1-score of 60%, and 38%, respectively. These scores correspond to performance of 39% and 15% above chance level, representing a good first step to the problem, given the difficulty of the dataset. Further, the training dataset and insights from this model provide a useful starting point for future work on the topic.