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

Abstract

Neutron–gamma discrimination plays a fundamental role in the fields of radiation protection and nuclide identification. In this paper, the neutron–gamma discrimination algorithm based on the combination of the kernel principal component analysis (KPCA) and the gaussian mixture model (GMM) was studied. The EJ309 liquid scintillation detector was used to detect two radioisotope neutron sources (252Cf and 241Am-Be). KPCA was used to reduce the dimension of the characteristic value of the pulse signal, GMM was used to cluster the dimensionality reduction results. The suitability of silhouette coefficient (S) as a discriminating index of traditional method and machine learning method was studied. The KPCA-GMM was compared with charge comparison method (CCM) and frequency gradient method (FGA) by using silhouette coefficient. The results show that the KPCA-GMM has higher discrimination accuracy. In the radiation field of 252Cf and 241Am-Be, the S of the KPCA-GMM reached 0.91 and 0.93, which were higher than those of CCM (0.86 and 0.86) and FGA (0.85 and 0.84). Therefore, the KPCA-GMM has good generalization ability and can be used for neutron–gamma discrimination.