Automatically identifying geologic features on planetary surfaces in remote sensing images using machine learning dramatically reduces the time, effort, and cost required for mapping. Often, creating a massive global map of features is not even feasible for human experts, as features can number in the hundred thousands to millions.
Localized studies have found populations of pitted cones and mounds in the northern hemisphere of Mars that may have been formed by subsurface ice or mud layers, indicating past aqueous saturation.
We use a U-Net architecture and train on local examples to map features globally at the highest resolution available for Mars (5m/pix).
Paper in progress (2023).