Parallel Factor Analysis and Support Vector Machines for Neutron-Gamma Discrimination

In order to perform a fast and accurate neutron-gamma discrimination, we present in this paper a method based on supervised and unsupervised machine learning that is composed of the following steps. Firstly, we apply nonnegative parallel factor analysis to recover the original sources from mixed signals recorded at the output of a stilbene scintillator detector (45×45 mm). Secondly, spectral analysis based on the continuous wavelet transform is used to characterize these recovered original sources. Thirdly, the resulting time-scale representations are considered as images that are processed using the Otsu segmentation method in order to get the binary images and thus extract attributes of interest of neutrons and gamma-rays signals from its background. Finally, we used principal component analysis to select the most significant of these attributes that are used as inputs of a support vector machines (SVM) to discriminate and classify the neutrons from gamma-rays. To evaluate the performance of the SVM model, bias-variance analysis is used. The results show that the proposed method can achieve an operational SVM prediction model for neutron-gamma classification with a high true positive rate.