Toward neutron / gamma discrimination with proportional counter using artificial intelligence

Extending the neutron detection capabilities of the ROSPEC SP2-1 proportional counter below 50 keV requires effective discrimination between neutron and gamma-ray signals at low energies. To address this challenge, we use a digital acquisition system, then apply signal processing and artificial intelligence techniques to analyze the data. The results showed that a CNN-based trained model can successfully distinguish noise from event signals in the recorded data. Key pulse features, such as rise time and amplitude, are extracted from the true signals to generate a two-dimensional plot of rise time versus amplitude, which facilitates the discrimination of neutron from gamma components. The application of the unsupervised clustering algorithm DBSCAN on this feature space shows limitations in accurately identifying low-amplitude gamma signals, while a measurement with only a gamma source confirms the presence of gamma events in the expected feature space. These results motivate the development of a supervised CNN-based approach to improve neutron/gamma discrimination.