Convolutional Neural Network-Based Neutron and Gamma Discrimination in EJ-276 for Low-Energy Detection

TL;DR

This paper demonstrates that a convolutional neural network significantly improves neutron and gamma discrimination in organic scintillators at low energies, achieving over 97% accuracy and outperforming traditional methods.

Contribution

The study introduces a CNN-based particle identification method that enhances low-energy neutron and gamma discrimination in EJ-276 scintillators, surpassing conventional techniques.

Findings

Achieved 97.3% accuracy in 0-100 keVee range.

Achieved 98.6% accuracy in 100-200 keVee range.

Improved discrimination by 13.8% and 4.25% over traditional methods.

Abstract

Organic scintillators are important in advancing nuclear detection and particle physics experiments. Achieving a high signal-to-noise ratio necessitates efficient pulse shape discrimination techniques to accurately distinguish between neutrons, gamma rays, and other particles within scintillator detectors. Although traditional charge comparison methods perform adequately for ~MeVee particles, their efficacy is significantly reduced in the lower energy region(<200 keVee). This paper introduces a particle identification method that harnesses the power of a convolutional neural network. We focused on the convolutional neural network's exceptional ability to discriminate between neutrons and gamma rays in the low-energy spectrum, utilizing a setup comprising a plastic scintillator EJ-276 and Silicon photomultiplier readout. Our findings reveal remarkable accuracies of 97.3% and 98.6% in the 0~100 keVee and 100~200 keVee energy ranges, respectively. These results represent substantial improvements of 13.8% and 4.25% over conventional methods. The enhanced discrimination power of the convolutional neural network method opens new frontiers for the application of organic scintillation detectors in low-energy rare event searches, including dark matter and neutrino detection.