Study on neutron–gamma discrimination method based on the KPCA-GMM-ANN

Abstract

Neutron–gamma discrimination plays a fundamental role in the fields of radiation protection and nuclide identification. The pulse-shape discrimination (PSD) of liquid scintillators has been widely used as an effective neutron–gamma detection method. In this paper, the KPCA-GMM-ANN combined algorithm is studied for neutron–gamma discrimination. Two radioisotope neutron sources (252Cf and 241Am–Be) were used. The EJ309 liquid scintillator detector was used to output the neutron and gamma pulse signals. The kernel principal component analysis (KPCA) was used to reduce the dimension of the characteristic value of the pulse signal, the gaussian mixture model (GMM) was used to cluster the dimensionality reduction results, and the artificial neural network (ANN) was trained to identify the particle categories and output the clustering results. The KPCA-GMM was compared with the KPCA-GMM-ANN using F1 score(F1). The discrimination error ratio (DER) was used to compare the KPCA-GMM-ANN with others' ANN. In all datasets, the KPCA-GMM-ANN's F1 were higher than 0.97, and its DER were lower than 0.035. The results show that the KPCA-GMM-ANN has good discrimination effect, can handle high-dimensional data, and has good generalization ability.