Deep learning on simulated gamma spectra for explosives detection using a NaI detector

Abstract

The detection of explosives and contraband materials using neutron activation analysis (NAA) is a critical component of modern security systems. This study investigates the feasibility of identifying explosive materials using a simple sodium iodide (NaI) scintillation detector limited to a 3 MeV gamma energy range. The detector’s limitations pose a significant challenge as characteristic gamma photopeaks above this range, such as those near 10 MeV, are excluded. Utilising a 14 MeV neutron source, gamma spectra from simulated neutron interactions with explosive materials were analysed using Geant4. This work demonstrates that with advanced machine learning models, such as convolutional neural networks (CNNs) and tailored data preprocessing methods, effective discrimination between explosives and non-explosives is achievable despite these constraints.