Title: Enhancing Low-Energy Neutron and Gamma Ray Detection Using Convolutional Neural Networks with EJ-276 Scintillators

Abstract: Organic scintillators, such as plastic scintillators, are widely used to detect fast neutrons and gamma rays. The EJ-276 scintillator offers a versatile solution for detecting fast neutrons and gamma rays simultaneously, making it ideal for mixed neutron-gamma field detection applications. This study evaluates the Pulse Shape Discrimination (PSD) capabilities of the EJ-276 scintillator paired with silicon photomultiplier (SiPM) array readouts. Integrating the 1-inch EJ-276 scintillator with SiPM arrays achieved a Figure of Merit (FOM) of 1.13 at an energy threshold of 200 keVee (electron equivalent). However, using the Charge Comparison Method (CCM) to distinguish between neutrons and gamma rays was challenging, especially at energies below 200 keVee. To improve low-energy resolution, the Convolutional Neural Network (CNN) approach was adopted. The InceptionTime and EfficientNetV2 models were developed, using one-dimensional time series and two-dimensional matrix image inputs, respectively. The transformation from one-dimensional arrays to two-dimensional images was achieved using three techniques: Gramian Angular Summation Field(GASF), Recurrence Plot(RP), and Relative Position Matrix(RPM). These methods demonstrated high accuracy at energy levels above 200 keVee. At lower energy regions, CNN methods, particularly the InceptionTime model, outperformed CCM methods. Notably, CNN methods reached accuracies of 96.79% and 98.33% in the 0-100 keVee and 100-200 keVee ranges, respectively, significantly higher than the 85.49% and 94.56% achieved by CCM, representing improvements of 13.22% and 3.99%. These results highlight the superior performance of CNN methods in differentiating between neutrons and gamma rays, especially in low-energy regions.