Logistic regression-time–frequency algorithm for α/β discrimination in silicon photomultiplier–based Phoswich detectors

Jiaqi Wang,Zhiyuan Li,Kun Wu,Jiaming Li,Zungang Wang,Jiangmei Zhang

DOI: https://doi.org/10.1016/j.nima.2024.169077

2024-01-24

Abstract:Phoswich (phosphor sandwich) systems composed of ZnS(Ag) and plastic dual scintillators have the ability to distinguish α and β particles via digital pulse-shape discrimination. In this work, a novel logistic regression algorithm based on time–frequency features (LR-TF) is presented and evaluated for its ability to digitally discriminate α and β particles in a mixed field. The LR-TF algorithm employs the charge comparison method (CCM) and Fourier gradient analysis (FGA) to extract time–frequency features from pulse shapes, thereby constructing a logistic regression model. Training of the model is shown to enable the discrimination α and β particles. The CCM, FGA, and LR-TF were applied to thousands of pulses obtained with a data acquisition system based on a Phoswich coupled with a silicon photomultiplier. The experimental results show that the LR-TF algorithm provided the best overall performance with misidentification ratios of 1.9 % and 0.63 % for α particles and β particles, respectively, achieving a figure of merit of 4.30, and an area under the receiver operating characteristic of 0.996.

physics, particles & fields, nuclear,nuclear science & technology,instruments & instrumentation