AI and the World's Hard Problems · Astronomy & space · arXiv · 2026-10-19
A New AI/ML Based Image Reconstruction Technique for Hard X-ray Coded Aperture Imagers
This work proposes a neural-network method that maps detector counts from a coded-aperture X-ray instrument directly to source positions, as an alternative to the standard FFT-based reconstruction. Trained on simulated responses from the Swift Burst Alert Telescope, the network combines convolution with attention. In simulation, it localized sources to the exact pixel over 93% of the time with 400 photons and no background, and kept at least 90% accuracy for 900-photon sources with up to 8,500 added background counts.