DoMars16k: A Diverse Dataset for Weakly Supervised Geomorphologic Analysis on Mars

Abstract

Mapping planetary surfaces is an intricate task that forms the basis for many geologic, geomorphologic, and geographic studies of planetary bodies. In this work, we present a method to automate a specific type of planetary mapping, geomorphic mapping, taking machine learning as a basis. Additionally, we introduce a novel dataset, termed DoMars16k, which contains 16,150 samples of fifteen different landforms commonly found on the Martian surface. We use a convolutional neural network to establish a relation between Mars Reconnaissance Orbiter Context Camera images and the landforms of the dataset. Afterwards, we employ a sliding-window approach in conjunction with a Markov Random field smoothing to create maps in a weakly supervised fashion. Finally, we provide encouraging results and carry out automated geomorphological analyses of Jezero crater, the Mars2020 landing site, and Oxia Planum, the prospective ExoMars landing site.

Keywords

• geomorphology
• Mars
• weak supervision
• mapping
• deep learning
• semantic segmentation