Development of an Artificial Neural Network for Special Nuclear Material Detection in a Mixed Photon-Neutron Environment

Abstract

Effective detection and characterization of special nuclear materials (SNMs) require state-of-the-art detectors and classification methods. The stilbene organic scintillator is an emerging radiation detector with the ability to simultaneously detect photons and neutrons. Dual-particle sensitivity presents a challenge in photon-induced active interrogation, where the photon flux is significantly higher than the neutron flux. Accurately detecting neutrons with minimum misclassification is crucial in these intense photon environments when looking for SNM, and the desired accuracy cannot be easily achieved using traditional charge integration (CI) methods. To address this need, an artificial neural network (ANN) system is developed to identify photon, neutron, and piled-up pulses; where possible, the piled-up pulses are recovered and classified accordingly. The developed ANN system is tested on time-of-flight (TOF) data collected at a variety of photomultiplier tube (PMT) gain settings. Across all gain settings, the ANN system produced similar classifications as the traditional CI method. The ANN performance is then evaluated in the presence of an intense photon field from a 9 MeV linear accelerator (linac). A 252Cf spontaneous fission source is measured in the presence of an intense photon environment. The ANN system demonstrated its robustness in an intense photon field by estimating the true particle count rates. However, the traditional CI method vastly overestimated neutron count rates in such intense photon field due to misclassification of pile-ups as neutrons.