Mapping linear surface features on Europa using a deep learning framework

The surface of Jupiter's icy moon Europa shows curvilinear geological features, so called lineaments. Some of them span over a hemisphere, while others appear only on a regional scale. These curvilinear surface features that potentially stem from cracks in the ice shell are of keen interest because they might provide a direct or indirect connection to Europa's subsurface ocean, allowing a remote sensing study of the subsurface ocean.
The solid-state imager onboard the Galileo mission observed Europa between 1996 and 2002 during 11 flybys and sent back data of almost 2 gigabyte. Based on a global map mosaicked from Galileo and Voyager images at a scale of 1:15M, Leonard et al. (2019) created a global map of the surface of Europa. Their mapping shows that ridged plains make up a major part of the surface area. Ridged plains are seemingly smooth but contain a high amount of undifferentiated lineae visible at higher resolution. 

We attempt to create a global map of lineaments at a higher resolution than the global geologic map. Although for the Galileo dataset, this mapping could be done manually, we need to prepare for a bigger data return by NASA's Europa Clipper mission. For this purpose, we introduce a deep learning framework that can map linear surface features in Galileo images on Europa autonomously and apply it on a global scale. More specifically, we train a Mask R-CNN that can detect, classify and segment lineaments. The current status of the work is presented.

References:
[1] Leonard, E. J., Senske, D. A., Patthoff, D. A., Global and Regional scale Geologic Mapping of Europa, EPSC-DPS2019-57-1, Vol. 13, 2019