Neutron-gamma discrimination based on support vector machine combined to nonnegative matrix factorization and continuous wavelet transform H. Arahmane a,⇑ , A. Mahmoudi b , E.-M. Hamzaoui c , Y. Ben Maissa d , R. Cherkaoui El Moursli a a ESMaR Laboratory, Faculty of Sciences, Mohammed V University, Rabat, Morocco b LIMIARF Laboratory, Faculty of Sciences, Mohammed V University, Rabat, Morocco c National Centre for Nuclear Energy, Science and Technology (CNESTEN), Rabat, Morocco d Laboratory of Telecommunication Systems, Networks and Services, National Institute of Posts and Telecommunications, Rabat, Morocco article info Article history: Received 3 November 2018 Received in revised form 11 August 2019 Accepted 15 August 2019 Available online 19 August 2019 Keywords: Neutron-gamma discrimination Nonnegative matrix factorization (NMF) Continuous wavelets transform (CWT) Time-scale representation Otsu global thresholding method Support Vector Machine (SVM) abstract Recent developments of digital signal processing have played an effective role to achieve a fast and accurate neutron-gamma discrimination. Thus, we present in this research work, a novel method which combines supervised and unsupervised machine learning to perform the neutron-gamma discrimination task at the output of a stilbene organic scintillation detector. We propose a three steps procedure that highly qualifies the discrimination. First, the detector’s output signals are processed as mixtures of several unknown sources, through nonnegative matrix factorization algorithms. Second, the continuous wavelet transform is performed to characterize the recovered original sources. The resulting time-scale representation is considered as an image which is segmented in order to extract main features of neutron signals versus the gamma ones. The features are then used as input of a nonlinear support vector machine classifier to finally achieve the neutron-gamma discrimination in mixed radiation field. Furthermore, the proposed method provides the classification precision for each radiation. Ó 2019 Elsevier Ltd. All rights reserved. 1. Introduction Neutron detection is greatly vital for nuclear research spec- troscopy. The scintillation detectors are the most adapted to this detection and preferred means for neutron spectroscopy purposes. However, their sensitivity to gamma-rays as well disturbs the esti- mation accuracies of the neutron pulses [1]. Due to their specific feature of organic scintillation and varying quality of discrimina- tion, pulse shape discrimination (PSD) methods were commonly used to perform the neutron-gamma discrimination [2,3]. The most popular ones are: charge comparison [4], rise time [5], zero-crossing [5] as PSD conventional methods and pulse gradient analysis (PGA) as PSD digital one [6]. Heltsley et al. have used charge comparison (or charge-integrating ADC) to sample the intensity in two different time regions of a pulse and thus to sense the shape of the pulse from a liquid scintillation detector [4]. Roush et al. have used rise time and zero-crossing methods to separate the scintillation signals from particles with different specific ionization based on signals produced by RC differentiation of dyn- ode voltage pulses [5]. Mellow et al. employed PGA approach that is based on the normalized amplitude of a single sample early in the decay of the pulse [6]. In addition, the wavelet transform is also appropriate for digital processing of a mixed radiation field. You- sefi et al., have been the first to propose it; converting the time to frequency domain pulses via a Fourier transform. The quality of neutron-gamma discrimination using these standard PSD meth- ods has conventionally been quantified by Figure of Merit (FOM) parameter defined as a distance between two pulses in a specific energy range, divided by the sum of the full-widths at half- maximum (FWHM) of the PSD ratio pulses. A high FOM is inter- preted as a good discriminator of the detector ability to discern the pulses generated by neutron and gamma radiations. However; in the presence of pulse pile-up; FOM is not able to well discrimi- nate the gamma contribution, and can be a little misleading [7]. Thus, we can say that the PSD methods are insufficient to perform the separation task with high precision. With improvements in programmable logic integrated circuits and in high-speed flash analog-to-digital converters, new possibil- ities in digital pulse processing (DSP) for scintillation detector systems based on digital event-by-event data acquisition have https://doi.org/10.1016/j.measurement.2019.106958 0263-2241/Ó 2019 Elsevier Ltd. All rights reserved. ⇑ Corresponding author. E-mail addresses: hanane_ar1@hotmail.com (H. Arahmane), benmaissa@inpt.ac. ma (Y. Ben Maissa). Measurement 149 (2020) 106958 Contents lists available at ScienceDirect Measurement journal homepage: www.elsevier.com/locate/measurement