Mapping Mars: Deep learning could help identify Jezero Crater landing site

Preparations for a secure touchdown on Earth, comparable to discovering essentially the most even terrain and equipping the suitable touchdown gear, are additionally essential for Mars missions.

Thus, touchdown a rover on Mars requires cautious mapping and planning nicely earlier than a rover’s descent begins. Scientists are working to create correct 3D floor maps, referred to as digital terrain fashions, of the planet by compiling mosaics of photographs from previous missions.

Progress in picture processing applied sciences over the previous 20 years has superior map resolutions from hundreds-of-meter to submeter scales. Though that is a unprecedented enchancment, even resolutions of 1 meter per pixel can’t absolutely seize fine-scale options like dune textures, small craters, and enormous rocks.

To raised map these geologic options across the 2020 Perseverance touchdown web site at Jezero crater, Yu Tao and colleagues use a deep learning model referred to as Multi-scale Generative Adversarial U-Internet (MADNet), which they designed in earlier work. The brand new analysis is published within the journal Earth and Area Science.

MADNet, educated utilizing a mix of present, postprocessed digital terrain fashions with resolutions starting from 4 to 36 meters per pixel, refined the publicly obtainable Mars 2020 Terrain Relative Navigation Excessive Decision Imaging Science Experiment (HiRISE) digital terrain model mosaic. The researchers additionally checked and refined a number of iterations to remove artifacts and gaps within the outputs.

The result’s the 50-centimeter-per-pixel MADNet HiRISE Jezero digital terrain mannequin mosaic. In contrast with the unique mosaics, the MADNet maps have a median top distinction of simply 0.009 meters, with a 0.63-meter standard deviation, indicating that the outcomes of the deep studying strategy align with the normal photogrammetric strategy.

The researchers observe that their product reveals important enhancements over present maps, together with (1) elevated efficient resolutions that present fine-scale floor options like dunes, craters, and rocks; (2) decreased striping artifacts; (3) the elimination of areas with low matching qualities; and (4) the elimination of interpolation artifacts. Their outcomes are publicly available.

This story is republished courtesy of Eos, hosted by the American Geophysical Union. Learn the unique story here.