An object detector adapted to damaged glass plates star and source markers to rescale their position on the sky, with a recovery rate of 97.3% and a two and a half times faster extraction.
Photographic plates accumulated by observatories for more than a century offer a valuable step back to study the variability of the sky. Still it is necessary to be able to rescale them precisely on the celestial coordinates, a step that wear, scratches and defects of the support often fail. Researchers present on arXiv Plate-YOLO, a detector that identifies sources and distinguishes stars in a single pass, before transmitting the catalogue obtained to an astrometric calibration software.
Refined with Gaia
The system relies on YOLOv5, modified to better identify very small objects. To adapt to a new collection, researchers refine it from the first resolute plates, whose labels are built using the Gaia satellite catalogue. Tested on 150 plates from the UBAI archive in Uzbekistan, on which the conventional processing chain had failed, the model dragged on Chinese plates in solution 40 immediately, and 106 more after refining, or 146 in total (97.3%).
The tool is also faster: 6.61 seconds per plate to extract the sources, compared to 16.73 seconds for SExtractor, the reference software, on the same machine.
Read the full article on the original source
Go to the source →