信号采样对Cs_(2) LiYCl_(6):Ce^(3+)探测器中子-伽马甄别性能的影响  

Effect of signal sampling on neutron-gamma discrimination performance of Cs_(2)LiYCl_(6)∶Ce^(3+)detector

作　　者：吴坤 黄广伟 王利斌 李林祥 席善学 陈声强 徐思 张立功 朱红英 王尊刚 刘辉兰[1] 宋玉收[1] 周春芝 WU Kun;HUANG Guangwei;WANG Libin;LI Linxiang;XI Shanxue;CHEN Shengqiang;XU Si;ZHANG Ligong;ZHU Hongying;WANG Zungang;LIU Huilan;SONG Yushou;ZHOU Chunzhi(College of Nuclear Science and Technology,Harbin Engineering University,Harbin 150001,China;State Key Laboratory of NBC Protection for Civilian,Academy of Military Sciences,Beijing 102205,China)

机构地区：[1]哈尔滨工程大学核科学与技术学院,黑龙江哈尔滨150001 [2]军事科学院防化研究院国民核生化灾害防护国家重点实验室,北京102205

出　　处：《哈尔滨工程大学学报》2022年第11期1670-1676,共7页Journal of Harbin Engineering University

基　　金：中央高校基本科研业务费项目(3072020CFT1505).

摘　　要：针对核脉冲信号数字处理中采样率与采样深度对粒子脉冲形状甄别效果的影响问题,本文基于Cs_(2)LiYCl_(6)∶Ce^(3+)(CLYC)探测器进行^(239)Pu-Be中子场中子/伽马射线混合信号采集,研究几种中子/伽马射线脉冲波形甄别算法以及这些算法对波形采样率和采样深度的适应性。研究结果表明:基于积分算法的电荷比较法与内积算法的k-means聚类配合向量投影法对采样率、采样深度、噪声的适应能力都好于基于微分算法的脉冲梯度法;由于k-means聚类配合向量投影法对向量维度敏感,在抗低采样率方面不如改进的电荷比较法,而在抗低采样深度方面二者能力相当;XGBoost和LightGBM这2种机器学习算法在采样率降至12.5 MS/s和采样深度降至4 bit后仍可获得100%的甄别准确度,在波形采样率和采样深度较低时相比传统算法优势明显。The sampling rate and depth in the digital processing of nuclear pulse signals have a considerable impact on the screening effect of particle pulse shape.The Cs_(2)LiYCl_(6)∶Ce^(3+)detector is used in this paper to collect neu-tron/gamma(γ)ray mixed signals in the ^(239)Pu-Be neutron field to study several neutron-gamma ray pulse waveform discrimination algorithms and their adaptability to the sampling rate and depth.Results show that the charge com-parison method based on integral operation and the k-means clustering combining vector projection method based on inner product operation adapts better to the sampling rate,sampling depth,and noise than the pulse gradient meth-od of differential operators.The k-means clustering combining vector projection method is sensitive to vector dimen-sions.Thus,this method is not as good as the improved charge comparison method in anti-low sampling rate,while the two methods have the same capacity in anti-low sampling depth.The two machine learning algorithms,XGBoost and LightGBM,can achieve 100%screening accuracy after the sampling rate is reduced to 12.5 MS/s and the sampling depth is decreased to 4 bits.Compared with traditional algorithms,the aforementioned algorithms have advantages in low waveform sampling rate and sampling depth.

关 键 词：Cs_(2)LiYCl_(6)∶Ce^(3+)探测器 中子/γ甄别 采样率 采样深度 电荷比较法 向量投影法 聚类 极端梯度提升机 轻量级梯度提升机 

分 类 号：TL812.1[核科学技术—核技术及应用]