Reactor-emitted electron antineutrinos can be detected via the inverse beta decay reaction, which produces a characteristic signal: a two-fold coincidence between a prompt positron event and a delayed neutron capture event within a specific time frame. While liquid scintillators are widely used for detecting neutrinos reacting with matter, detection is difficult because of the low interaction of neutrinos. In particular, it is important to distinguish between neutron (n) and gamma (γ) signals. The principle of the interaction of neutrons with matter differs from that of gamma rays with matter, and hence the detection signal's waveform is different. Conventionally, pulse shape discrimination (PSD) is used for n/γ separation. This study developed a machine learning method to see if it is more efficient than the traditional PSD method. The possibility of n/γ discrimination in the region beyond the linear response limits was also examined, by using 10- and 2-inch photomultiplier tubes (PMTs) simultaneously. To the best of our knowledge, no study has attempted PSD in a PMT nonlinear region using artificial neural networks. Our results indicate that the proposed method has the potential to distinguish between n and γ signals in a nonlinear region.
Keywords: artificial neural network; gamma and neutron separation; linearity; liquid scintillator; photomultiplier tube; pulse shape discrimination; saturation.