Studies of Gamma-Ray Shower Reconstruction Using Deep Learning

TL;DR

This paper explores the use of convolutional neural networks to improve the reconstruction of gamma-ray events at low energies in cosmic ray detection, aiming to outperform traditional analysis methods.

Contribution

It introduces a CNN-based approach for gamma-ray shower reconstruction using surface array data, focusing on low-energy events below 1 TeV, which is a novel application in this context.

Findings

CNNs improve reconstruction accuracy at low energies

Deep learning outperforms traditional SEMLA analysis

Enhanced detection capabilities for soft-spectrum gamma-ray sources

Abstract

The Cosmic Multiperspective Event Tracker (CoMET) R&D project aims to optimize the techniques for the detection of soft-spectrum sources through very-high-energy gamma-ray observations using particle detectors (called ALTO detectors), and atmospheric Cherenkov light collectors (called CLiC detectors). The accurate reconstruction of the energies and maximum depths of gamma-ray events using a surface array only, is an especially challenging problem at low energies, and the focus of the project. In this contribution, we leverage Convolutional Neural Networks (CNNs) using the ALTO detectors only, to try to improve reconstruction performance at lower energies ( < 1 TeV ) as compared to the SEMLA analysis procedure, which is a more traditional method using manually derived features.