Deep learning maps of fresh crater ejecta on Mercury

Data used in the work "Deep learning map of fresh crater ejecta on Mercury" to create a map of the immature ejecta of fresh impact craters on Mercury using a deep learning method: EJMAP. Contents ejecta_map_masked.tif: the main data product, a global 665-m fresh crater ejecta map in Equidistant Cylindrical projection containing the following bands. EJMAP ejecta maps. Ejecta are set to the identifying code (FID) of their progenitor crater (see fresh_crater.zip). Non-ejecta pixels are set to -1, null pixels to -2, overlaps between ejecta to -3. EJMAP quality mask. Legend: 0: non-ejecta pixels. 1: ejecta blanket pixels from single crater. 2: ejecta ray pixels from single crater. 10: ejecta overlaps between multiple craters, exclusion recommended to mitigate map errors and avoid false positive. 11: zones within two crater radii from the center, exclusion recommended to avoid ejecta with distinct compositional signature. 12: other ejecta pixels whose exclusion is recommended. 13: null pixels Manual ejecta maps. Include manual maps of the 10 manually-mapped MAN craters and the manually-mapped ejecta of the other craters in the training tiles. Same legend as Band 1. Manual maps quality mask. Same legend as Band 2. The blanket radii are recalculated on the MAN maps, except for Hokusai, while for the other craters they are the same as the EJMAP radii. fresh_craters.zip : the shapefile of the 284 selected fresh craters, containing their outlines, FIDs and geometry information final_ejecta_masks.zip : the ejecta raster masks of the individual craters produced by EJMAP. manual_masks.zip : the ejecta masks of the 10 manually-mapped MAN craters (in azimuthal projection centered on the crater). results.zip : the performance metrics of the various model versions. For the final models (conn_results/version_32; ejc_results/version_200), the model weights are also included. Download and unzip this folder in the directory structure illustrated in the code repository. data.zip : the data necessary for replication, including the manual ejecta masks for the tiles used to train the deep learning algorithm. Download and unzip this folder in the directory structure illustrated in the code repository.