A Neural-Network Framework for Tracking and Identification of Cosmic-Ray Nuclei in the RadMap Telescope

2025/08/18 by Meyer-Hetling, Luise, Losekamm, Martin J., Paul, Stephan +1
#FOS: Physical sciences #Instrumentation and Detectors (physics.ins-det) #Instrumentation and Methods for Astrophysics (astro-ph.IM) #Space Physics (physics.space-ph)

paper · doi:10.48550/arxiv.2508.12708

Abstract

We present a neural-network framework designed to reconstruct the properties of cosmic-ray nuclei traversing the scintillating-fiber tracking calorimeter of the RadMap Telescope. Employing the Geant4 simulation toolkit and a simplified model of the detector to generate training and test data, we achieve the spectroscopic capabilities required for an accurate determination of the biologically relevant dose that astronauts receive in space. We can reconstruct a particle's trajectory with an angular resolution of better than 1.4^∘ and achieve a charge separation of better than 95% for nuclei with Z≤8; specifically, we reach an accuracy of 99.8% for hydrogen. The energy resolution is <20% for energies below 1 GeV/n and elements up to iron. We also discuss the limitations of our detector, the reconstruction framework, and this feasibility study, as well as possible improvements.

Citations

• Attention Is All You Need
• Going Deeper with Convolutions
• The Graph Neural Network Model
• Gradient-based learning applied to document recognition

Related

• Neural Networks as a Composition Diagnostic for Ultra-high Energy Cosmic Rays
• Particle identification in ground-based gamma-ray astronomy using convolutional neural networks
• Neural Network Constraints on the Cosmic-Ray Ionization Rate and Other Physical Conditions in NGC 253 with ALCHEMI Measurements of HCN and HNC
• Position reconstruction using deep learning for the HERD PSD beam test
• Inverting cosmic ray propagation by Convolutional Neural Networks