Effective pulse shape discrimination for neutron and gamma-ray radiation using wavelet transform and logistic regression

Abstract

Abstract Accurate classification of neutron and gamma radiation remains a fundamental challenge in radiation detection due to their overlapping pulse shape characteristics in scintillator detectors. This paper presents a novel machine learning approach for pulse shape discrimination that combines effective signal processing with bio-inspired optimization. Our methodology employs wavelet decomposition to extract discriminative time-frequency features from radiation pulses, which are then classified using a probabilistic logistic regression model. To ensure robust training, the charge comparison method is utilized as a pre-selection process that provides high-purity neutron and gamma pulse datasets. The system is further refined through Artificial Bee Colony optimization, which automatically tunes critical parameters via a custom fitness function designed to maximize the achievable discrimination performance. The effectiveness of the proposed work is illustrated through a performance comparison with traditional and stat-of-the-art discrimination techniques. Experimental results show that the proposed method outperforms other approaches in terms of the figure of merit that quantifies the discrimination capability.