Robust pulse shape discrimination algorithm using canonical correlation analysis of gamma and neutron pulses

In this paper, a robust pulse shape discrimination algorithm is proposed to classify neutron and gamma pulses in radiation detectors. Scalogram of the induced pulses, which represent the absolute value of the Continuous Wavelet Transform (CWT) coefficients in the time-frequency domain, is used as feature descriptors. A two-dimensional principal component analysis (2D PCA) is employed to achieve dimensionality reduction of the obtained features and remove the redundancy in the given data. Accordingly, dominant features are selected and stacked together to construct two feature sets representing the two classes (neutron and gamma). Canonical Correlation Analysis is performed between the testing and training feature sets to find the basis vectors on which the new projected set pairs are highly correlated. Experimental results showed that the proposed method outperforms the other state-of-the art PSD methods in terms the discrimination accuracy.