Separation of track- and shower-like energy deposits in ProtoDUNE-SP using a convolutional neural network

Abstract

Liquid argon time projection chamber detector technology provides high spatial and calorimetric resolutions on the charged particles traversing liquid argon. As a result, the technology has been used in a number of recent neutrino experiments, and is the technology of choice for the Deep Underground Neutrino Experiment (DUNE). In order to perform high precision measurements of neutrinos in the detector, final state particles need to be effectively identified, and their energy accurately reconstructed. This article proposes an algorithm based on a convolutional neural network to perform the classification of energy deposits and reconstructed particles as track-like or arising from electromagnetic cascades. Results from testing the algorithm on data from ProtoDUNE-SP, a prototype of the DUNE far detector, are presented. The network identifies track- and shower-like particles, as well as Michel electrons, with high efficiency. The performance of the algorithm is consistent between data and simulation.

31 pages, 15 figures

References in corpus (6)

• First results on ProtoDUNE-SP liquid argon time projection chamber performance from a beam test at the CERN Neutrino Platform
• A Convolutional Neural Network based Cascade Reconstruction for the IceCube Neutrino Observatory
• Neutrino interaction classification with a convolutional neural network in the DUNE far detector
• GPU-accelerated machine learning inference as a service for computing in neutrino experiments
• Space Charge Effects in Noble Liquid Calorimeters and Time Projection Chambers
• Reconstruction of interactions in the ProtoDUNE-SP detector with Pandora

Cited by in corpus (5)

• Self-compensating Light Calorimetry with Liquid Argon Time Projection Chamber for GeV Neutrino Physics
• Accelerating Machine Learning Inference with GPUs in ProtoDUNE Data Processing
• Graph Neural Network for Neutrino Physics Event Reconstruction
• Identification and reconstruction of low-energy electrons in the ProtoDUNE-SP detector
• LArTPC hit-based topology classification with quantum machine learning and symmetry