Detection of Geological Features on Terrestrial Planets by using Machine Learning

Understanding the geologic evolution of terrestrial planets in the Solar System is one of the most challenging objectives that space exploration missions have been attempting to accomplish. Deep space probes have unveiled surface features that might have originated from exogenic and endogenic mechanisms, including volcanism and tectonism. A better understanding of the formation of celestial bodies' surface is linked to our knowledge of similar processes on Earth. Thorough observations of these features yield fundamental constraints on planetary evolution.

This work focuses on the development and implementation of an automatic feature-detection algorithm that identifies volcano-tectonic morphologies on the most Earth-like planets in the Solar System, including Venus and Mars. Space missions (e.g., Magellan and Mars Reconnaissance Orbiter) provided Synthetic Aperture Radar (SAR) and optical imagery measurements of planetary surfaces, and incrementing data volume is expected from future missions to fulfill high-resolution requirements of scientific payloads (e.g., EnVision VenSAR).

Our solution is based on a Deep Convolutional Neural Network algorithm devoted to the task of classification and segmentation of SAR and optical images. To enable a more efficient data processing, our software is conceived to analyze large-size image datasets and to unambiguously detect volcano-tectonic features preventing from time consuming manual classification. The training is performed by using labels retrieved from accurate geological maps, available on the USGS repository in a GIS-compatible format.

The proposed approach will be used to enable an autonomous detection and characterization of features on planetary surfaces. Since an accurate mapping of geological features with orbital data is fundamental to determine region of interest for science investigations, the implementation of this technique will also be well-suited to support spacecraft navigation working with limited computation capability. Furthermore, an autonomous retrieval of the surface properties is also a critical task that guarantees safe operations during high-risk mission phases, including landing.