Abstract

Poster - Splinter MeVSky   (MW-0337)

CNNCat: Categorizing high-energy photons in a Compton/Pair Telescope with Convolutional Neural Networks

Jan Peter Lommler, Uwe Oberlack
Johannes Gutenberg-Universität Mainz

Low to medium energy gamma rays are shielded by the Earth's atmosphere and cannot be measured with on-ground facilities. Satellite based gamma-ray astronomy relies on Compton scatter and Pair creation as measurement channels. One of the biggest challenges is the poor signal to background ratio due to low signal fluxes from cosmic sources and the high background rates even in the comparatively moderate environment of Low Earth Orbits. An efficient event tagging reduces signal losses by preventing type-mismatching applications of reconstruction algorithms (e.g. performing a Compton reconstruction on a Pair event) and signal pollution (distinguishing events originating from background sources). We report results on the feasibility of Deep Convolutional Neural Nets in the context of event classification for Compton-Pair telescopes on the example of the e-ASTROGAM design proposal and give an outlook on applications to future telescopes.