Geology is a complex field and planetary geology even more so - no wonder that NASA and ESA are increasingly relying on the use of artificial intelligence. However, training a good AI algorithm needs a lot of data, which is not always available, or only to a very limited extent, due to the accessibility of the objects in the run-up to a space mission, as there is no or not enough "ground truth". Therefore, VRVis supports planetary research by artificially generating the data necessary for training neural networks.
Based on high-resolution 3D scans of terrestrial shatter cones (fan-shaped structures with grooves that result when a meteorite strikes), we generated exactly these rock formations artificially for the Martian surface. For this purpose, we embedded randomly shatter cones in the terrain model of Mars. Shading, coloring, scaling, and even overlaps were simulated photorealistically. The networks trained on this are an important contribution to making the work of geological science teams easier in the future by enabling them to sift through the ever-increasing number of images more efficiently and thus find interesting geological sites more quickly.
The talk of Chris Traxler starts at 1h 23min 56sec. Chris Traxler gave the talk "Visualization and Simulation for Artificial Intelligence (AI) Training in Space Exploration" at the ÖAW Symposium "The Mars Rovers and Austria" on 9 Decemeber 2020.
