Machine Learning Approach to Crater Mapping on Enceladus

This project aims to assess the applicability of a U-net Convolutional Neural Network (CNN) for crater detection on Saturn's moon Enceladus, trained on data from Cassini's Imaging Science Subsystem (ISS). So far, the development of similar ML models has been restricted to rocky, inner solar system bodies (i.e., Mars & the Moon), and have primarily made use of digital elevation data. Given the morphological differences between these bodies and the icy moons found in the outer solar system, as well as limitations DEMs present in both resolution and availability, the development of an image-based crater detection model for such bodies should be a high value task in the field of Planetary Science.

We used light-shadow based labels to train the CNN, allowing the final product to provide insights into not only crater distribution but also morphology.

Detailed mapping of the relaxed crater's distribution could give insights into heat flow variations on the body's surface, allowing for a more in-depth analysis of the crust's viscosity. In addition, studying crater morphology can help us understand multiple physical properties of the target body such as surface age, heat flow, and rheology. We suggest a collaboration between this machine learning approach and human reviewers for optimal mapping results.