Speaker
Description
Single muons traversing the field of view of an Imaging Atmospheric Cherenkov Telescope (IACT) produce rings of Cherenkov light whose geometry is fixed by the Cherenkov emission angle and the muon impact parameter. Because the expected light yield can be computed from first principles, muon rings act both as a standard candle for the instrument and as a direct probe of the muons’ properties. They are routinely used to monitor mirror reflectivity, to track the point-spread function and to provide an absolute calibration of the camera, while the distribution of ring radii constrains the muon energy spectrum, and the overall muon rate together with the lateral muon distribution constrains hadronic interaction models. Building a clean and well-characterised muon sample for these applications is, however, difficult: a substantial fraction of rings are partial or off-axis, and events combining a muon with a coincident air-shower image are common, so conventional cut-based selections struggle to make full use of the available muon events. We present a deep-learning approach to muon identification in IACT camera images that operates directly on the native camera pixel geometry, with modern neural architectures explored in supervised, unsupervised and self-supervised settings and benchmarked against classical baselines.