https://solarsystem.video/videos/watch/1e52c389-3926-4616-9565-11057f73308a Presented by Michele Lissoni (LIRA, Paris Observatory) ML4PSP Seminar October 2026 Crater rays are among the most prominent features on the surface of Mercury. Associated with the most recent impact craters, they are composed of high-albedo, immature ejecta, which are progressively darkened by space weathering. Distinguishing the ejecta patterns of craters formed at different times can therefore reveal how space weathering alters the planetary surface. Here, a deep learning algorithm, EJMAP, has been deployed to map the ejecta of 284 fresh craters using a 3-band global mosaic of Mercury and distinguish their crater of origin. EJMAP is composed of two Convolutional Neural Networks: EJCONN, a binary classification model determines whether a tile of the mosaic contains ejecta from a given crater, and EJSEG, a semantic segmentation model that maps said ejecta. In terms of F1-score, the performance of EJMAP is on average 0.6 at the pixel level and 0.76 at the tile level, but has been observed to vary widely depending on how training and test data are partitioned. The ejecta maps improve however closer to the craters, making it possible to study the spectral properties of the closest and brightest immature material. Thu, 22 Oct 2026 20:29:18 GMT https://validator.w3.org/feed/docs/rss2.html PeerTube - https://solarsystem.video https://solarsystem.video/client/assets/images/icons/icon-1500x1500.png https://solarsystem.video/videos/watch/1e52c389-3926-4616-9565-11057f73308a All rights reserved, unless otherwise specified in the terms specified at https://solarsystem.video/about and potential licenses granted by each content's rightholder.