Pulse shape discrimination using a convolutional neural network for organic liquid scintillator signals

TL;DR

This paper presents a CNN-based method for pulse shape discrimination in organic liquid scintillators, significantly improving neutron background rejection in neutrino detection experiments.

Contribution

The study introduces a novel CNN architecture that enhances PSD performance over traditional methods using Fourier-transformed waveforms.

Findings

Over 20% improvement in signal-to-background ratio in 1-10 MeV range

Greater enhancement observed at low energies

CNN effectively distinguishes beta and alpha events

Abstract

A convolutional neural network (CNN) architecture is developed to improve the pulse shape discrimination (PSD) power of the gadolinium-loaded organic liquid scintillation detector to reduce the fast neutron background in the inverse beta decay candidate events of the NEOS-II data. A power spectrum of an event is constructed using a fast Fourier transform of the time domain raw waveforms and put into CNN. An early data set is evaluated by CNN after it is trained using low energy $\beta$ and $\alpha$ events. The signal-to-background ratio averaged over 1-10 MeV visible energy range is enhanced by more than 20% in the result of the CNN method compared to that of an existing conventional PSD method, and the improvement is even higher in the low energy region.