Joint machine learning and analytic track reconstruction for X-ray polarimetry with gas pixel detectors

¹ Istituto Nazionale di Fisica Nucleare, Sezione di Torino, Via Pietro Giuria 1, 10125 Torino, Italy
e-mail: nicolo.cibrario@unito.it
² Dipartimento di Fisica, Università degli Studi di Torino, Via Pietro Giuria 1, 10125 Torino, Italy
³ University of Maryland, Baltimore County, Baltimore, MD 21250, USA
⁴ NASA Goddard Space Flight Center, Greenbelt, 8800 Greenbelt Rd, MD 20771, USA
⁵ Center for Research and Exploration in Space Science and Technology, NASA/GSFC, Greenbelt, 8800 Greenbelt Rd, MD 20771, USA
⁶ Department of Radiation Oncology, Netherlands Cancer Institute, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands
⁷ Istituto Nazionale di Fisica Nucleare, Sezione di Pisa, Largo B. Pontecorvo 3, 56127 Pisa, Italy
⁸ Dipartimento di Fisica, Università di Pisa, Largo B. Pontecorvo 3, 56127 Pisa, Italy
⁹ Department of Physics and Kavli Institute for Particle Astrophysics and Cosmology, Stanford University, 450 Serra Mall, Stanford, CA 94305, USA
¹⁰ Istituto Nazionale di Fisica Nucleare, Sezione di Napoli, Strada Comunale Cinthia, 80126 Napoli, Italy

Received: 2 March 2023
Accepted: 18 April 2023

Abstract

We present our study on the reconstruction of photoelectron tracks in gas pixel detectors used for astrophysical X-ray polarimetry. Our work aims to maximize the performance of convolutional neural networks (CNNs) to predict the impact point of incoming X-rays from the image of the photoelectron track. A very high precision in the reconstruction of the impact point position is achieved thanks to the introduction of an artificial sharpening process of the images. We find that providing the CNN-predicted impact point as input to the state-of-the-art analytic analysis improves the modulation factor (~1% at 3 keV and ~6% at 6 keV) and naturally mitigates a subtle effect appearing in polarization measurements of bright extended sources known as “polarization leakage”.