Fast neutron-gamma discrimination in organic scintillators via convolution neural network

Due to the high gamma sensitivity of organic scintillators, it is essential to discriminate signals induced by neutron from gamma-ray in fast-neutron detection. With the improvement of digital signal processing techniques, diverse discrimination methods based on pulse-shape variation by radiation type have been developed. The main purpose of this study was to verify the applicability of a deep-learning model, especially convolution neural network (CNN), to pulse-shape discrimination (PSD) in organic scintillation detectors, such as BC-501A (liquid) and EJ-276 (plastic). To that end, waveforms of neutron and gamma-ray were experimentally collected using point sources of 137 Cs (gamma-ray) and 252 Cf (neutron/gamma-ray) and pre-processed for being compatible with deep-learning. The PSD performance was evaluated for both detectors using the charge comparison method (CCM) which is one of the representative conventional PSD techniques of time-domain. In addition, the CNN-based discriminating algorithms were tested, and its preliminary results were confirmed with confusion matrices which indicate the discrimination accuracy of a deep-learning model.