Speaker
Description
Precisely identifying the astrophysical sources of cosmic rays remains challenging. Further improvements in the sensitivity to gamma-ray sources are therefore required. We present a Transformer-based deep learning model that uses two separate branches to extract features from heterogeneous data recorded by the electromagnetic particle detectors (EDs) and muon detectors (MDs) of LHAASO-KM2A, and then performs physics-informed feature fusion for gamma/hadron separation and direction reconstruction. Using six months of observations of the Crab Nebula, this method improves the gamma-ray signal significance of the Crab Nebula by more than 50% in the 10–25 TeV band and by more than 35% in the 25–100 TeV band relative to traditional methods, while also improving the angular resolution for gamma-ray sources.