Enhancing Neutron/Gamma Discrimination in the Low-Energy Region for EJ-276 Plastic Scintillation Detector Using Machine Learning

Pulse Shape Discrimination (PSD) techniques, particularly the widely employed charge integration ratio method (Q-ratio), have proven effective in discriminating fast neutrons from gamma rays in organic scintillation detectors. However, the effectiveness of Q-ratio diminishes in the low-energy region (below 150 keVee) due to overlapping signal, leading to a suboptimal Figure of Merit (FOM). In this study, we use machine learning (ML) technique, particularly the one-dimensional Convolutional Neural Network (1D-CNN), to enhance the neutron/gamma discrimination and compares the results with the traditional charge integration ratio in the low-energy region. Our investigation focuses on the EJ-276 plastic scintillator, a commercial product of ELJEN technology known for its good separation of gamma and fast neutron signals based on timing characteristics. Experimental data were acquired using
²⁵²
Cf and
⁶⁰
Co radioisotope sources. A comprehensive comparative analysis between the traditional Q-ratio method and ML algorithms is conducted for the low energy region. Our main objective is to evaluate and enhance neutron/gamma discrimination capabilities of plastic scintillators in this low-energy region.