LineaMapper: A deep learning-powered tool for mapping linear surface features on Europa

Haslebacher, Caroline; Thomas, Nicolas; Bickel, Valentin T. (2024). LineaMapper: A deep learning-powered tool for mapping linear surface features on Europa. Icarus, 410 Elsevier 10.1016/j.icarus.2023.115722

Preview

Text 1-s2.0-S0019103523002993-main.pdf - Published Version Available under License Creative Commons: Attribution (CC-BY). Download (10MB) | Preview

As discontinuities of the smooth icy surface, linear surface features might be directly or indirectly linked to Europa’s subsurface ocean. Mapping and categorising Europa’s lineaments is a means of retrieving information that could be linked to their formation history. As of today, planetary mapping is mainly conducted manually, which is tedious and subject to human bias once data sets become large. Mapping is further complicated by the heterogeneous quality and coverage of the available image data.

Here, we train LineaMapper, a convolutional neural network (Mask R-CNN), to conduct instance segmentation of the four main units of linear surface features on Europa: bands, double ridges, ridge complexes and undifferentiated lineae. LineaMapper is trained on the basis of 15 mosaics from the Galileo solid-state imager data, yielding 930 training tiles. With LineaMapper, we provide a new method that facilitates detailed mapping of lineaments in Galileo images. LineaMapper could be applied to data to be returned by the Europa Imaging System (EIS) onboard the Europa Clipper mission. We validate LineaMapper v1.0 on an independent test set. On this test set, LineaMapper shows an overall higher precision than recall. In other words, there are more non-detections of actual lineaments than there are false detections of lineaments. The model shows the most correct predictions for double ridges (highest precision), while the most complete detections happen for ridge complexes (highest recall), compared with the ground truth. In some cases, LineaMapper preserves the cross-cutting relationships. The biggest strength of LineaMapper lies in its speed and tunable output. In the future, LineaMapper can be retrained, fine tuned and applied to similar looking features, for example wrinkle ridges on Venus, ridges on other planets and moons or even dust devil tracks on Mars.

Interest & Impact

Downloads

1 since deposited on 02 Apr 2024
1 in the past 12 months

Citations

5 Citations in Web of Science ®
6 Citations in Scopus

Search

in Google Scholar™

Services

Actions (login required)

Edit item

Actions (login required)

Edit item