1FDL: the first catalogue of gamma sources from all over the sky developed by deep learning over ten years of Fermi data

More than 3,800 sources detected, located and recorded by neuron networks, with comparable or better results than the official chain when the galactic emission model varies.

An international team has adapted AutoSource-ID, a detection tool first designed for optical imaging, to observations from the Fermi-LAT Space Telescope. A U-Net segments gamma-ray maps of the sky to locate sources, then dedicated networks estimate their position and flux and decide whether it is a real source or an artifact. The training is based on Monte Carlo simulations modeled on ten years of observations.

The result, dubbed 1FDL, lists more than 3,800 sources, each with a probability of being authentic. It largely agrees with the 4FGL-DR2 reference catalog covering the same period, and even contains a few sources that only the recent 16-year list (FL16Y) has confirmed.

A strong point highlighted: in the face of uncertainties about the interstellar emission model, which weighs heavily on the detection of weak sources, the method is at least as robust as the plausibility approach used for official catalogues.