Augmenting astronomical X-ray detectors with AI for enhanced sensitivity and reduced background

TL;DR

Integrating AI with next-generation X-ray detectors significantly improves sensitivity and background rejection, enabling better detection of faint and distant astronomical sources.

Contribution

This work introduces prototype machine learning algorithms for background identification and demonstrates their effectiveness in reducing instrumental noise in X-ray detectors.

Findings

AI reduces instrumental background by up to 41.5%

Machine learning improves event classification accuracy

Enhanced spectral resolution for low-energy photons

Abstract

Bringing artificial intelligence (AI) alongside next-generation X-ray imaging detectors, including CCDs and DEPFET sensors, enhances their sensitivity to achieve many of the flagship science cases targeted by future X-ray observatories, based upon low surface brightness and high redshift sources. Machine learning algorithms operating on the raw frame-level data provide enhanced identification of background vs. astrophysical X-ray events, by considering all of the signals in the context within which they appear within each frame. We have developed prototype machine learning algorithms to identify valid X-ray and cosmic-ray induced background events, trained and tested upon a suite of realistic end-to-end simulations that trace the interaction of cosmic ray particles and their secondaries through the spacecraft and detector. These algorithms demonstrate that AI can reduce the unrejected instrumental background by up to 41.5 per cent compared with traditional filtering methods. Alongside AI algorithms to reduce the instrumental background, next-generation event reconstruction methods, based upon fitting physically-motivated Gaussian models of the charge clouds produced by events within the detector, promise increased accuracy and spectral resolution of the lowest energy photon events.