Identifying charged particle background events in X-ray imaging detectors with novel machine learning algorithms

Abstract

Space-based X-ray detectors are subject to significant fluxes of charged particles in orbit, notably energetic cosmic ray protons, contributing a significant background. We develop novel machine learning algorithms to detect charged particle events in next-generation X-ray CCDs and DEPFET detectors, with initial studies focusing on the Athena Wide Field Imager (WFI) DEPFET detector. We train and test a prototype convolutional neural network algorithm and find that charged particle and X-ray events are identified with a high degree of accuracy, exploiting correlations between pixels to improve performance over existing event detection algorithms. 99 per cent of frames containing a cosmic ray are identified and the neural network is able to correctly identify up to 40 per cent of the cosmic rays that are missed by current event classification criteria, showing potential to significantly reduce the instrumental background, and unlock the full scientific potential of future X-ray missions such as Athena, Lynx and AXIS.

A Probabilistic Method of Background Removal for High Energy Astrophysics Data

“…Recent work (Wilkins et al 2020;Grant et al 2020) using simulations of the upcoming ATHENA mission has shown that machine learning methods may be able to independently estimate the probability that an individual event is associated with the background. The methods presented here can be considered complimentary to those, in the sense that the our framework can be used in situations where such high fidelity simulations of X-ray photon interactions with telescope hardware are not available or feasible.…”