A Self-Organizing feature Map neural network for compton suppression in phoswich detector

Abstract

Scintillation phoswich detectors can be used for particle discrimination and Compton suppression, and their function is based on pulse shape discrimination (PSD) methods. The conventional PSD methods mainly including Rise Time Discrimination (RTD), Rise and Decay time discrimination (R&D), Constant Time Discrimination (CTD), etc., but linear PSD methods have their limitations in processing pulse signals. Neural network tools can achieve nonlinear output by adjusting the weights between neurons during the training, which are suitable for identifying nonlinear signals generated by detectors. The self-organizing feature mapping (SOM) is the most commonly used clustering neural network, which can cluster signals based on their similarity through competitive training of neural networks without prior data. In this study, 3 SOM networks with different size are trained, and after comparison with the single LaBr3:Ce detector, the total accuracy of the phoswich detector using the SOM neural network for discrimination reaches over 99.5%, which is a 7.59% increase in accuracy compared to the RTD method, and the peak-to-total ratio (P/T) is improved by 1.07%.