Uchida Y, Takada E, Fujisaki A, Kikuchi T, Ogawa K, Isobe M
National Institute of Technology, Toyama College, 13 Hongo-mach, Toyama 939-8630, Japan.
Nagaoka University of Technology, 1603-1 Kamitomioka, Niigata 940-2188, Japan.
Rev Sci Instrum. 2017 Aug;88(8):083504. doi: 10.1063/1.4996177.
A method to stochastically discriminate neutron and γ-ray signals measured with a stilbene organic scintillator is proposed. Each pulse signal was stochastically categorized into two groups: neutron and γ-ray. In previous work, the Expectation Maximization (EM) algorithm was used with the assumption that the measured data followed a Gaussian mixture distribution. It was shown that probabilistic discrimination between these groups is possible. Moreover, by setting the initial parameters for the Gaussian mixture distribution with a k-means algorithm, the possibility of automatic discrimination was demonstrated. In this study, the Student's t-mixture distribution was used as a probabilistic distribution with the EM algorithm to improve the robustness against the effect of outliers caused by pileup of the signals. To validate the proposed method, the figures of merit (FOMs) were compared for the EM algorithm assuming a t-mixture distribution and a Gaussian mixture distribution. The t-mixture distribution resulted in an improvement of the FOMs compared with the Gaussian mixture distribution. The proposed data processing technique is a promising tool not only for neutron and γ-ray discrimination in fusion experiments but also in other fields, for example, homeland security, cancer therapy with high energy particles, nuclear reactor decommissioning, pattern recognition, and so on.
提出了一种利用芪有机闪烁体测量随机区分中子和γ射线信号的方法。每个脉冲信号被随机分为两组：中子和γ射线。在之前的工作中，期望最大化（EM）算法是在测量数据遵循高斯混合分布的假设下使用的。结果表明，这些组之间的概率区分是可能的。此外，通过使用k均值算法设置高斯混合分布的初始参数，证明了自动区分的可能性。在本研究中，学生t混合分布被用作EM算法的概率分布，以提高对信号堆积引起的异常值影响的鲁棒性。为了验证所提出的方法，比较了假设t混合分布和高斯混合分布的EM算法的品质因数（FOM）。与高斯混合分布相比，t混合分布导致FOM有所改善。所提出的数据处理技术不仅是聚变实验中中子和γ射线区分的有前途的工具，而且在其他领域，例如国土安全、高能粒子癌症治疗、核反应堆退役、模式识别等方面也是如此。
