液体闪烁体探测器无监督n/γ甄别算法研究  

Research on Unsupervised n/γ Discrimination Algorithm Based on Liquid Scintillator Detector

作　　者：刘路峰 周红召 邵晖 LIU Lu-feng;ZHOU Hong-zhao;SHAO Hui(State Key Laboratory of NBC Protection for Civilian,Beijing 102205,China)

机构地区：[1]国民核生化灾害防护国家重点实验室,北京102205

出　　处：《核电子学与探测技术》2021年第6期925-931,共7页Nuclear Electronics & Detection Technology

基　　金：国家自然科学基金面上项目(12075318)。

摘　　要：利用液体闪烁体探测器对^(252)Cf、^(241)Am-Be和^(137)Cs进行探测。分析比较了主成分分析、核主成分分析和局部线性嵌入在数据处理中的降维效果。利用K-Means聚类算法、高斯混合模型和DBSCAN聚类算法对降维结果进行n/γ甄别。分析了FOM值在一些应用场景中的不足,并通过引用精度、召回率和F_(1)分数评估算法甄别效果。综合评估表明:核主成分分析与K-means组合的识别效果最好,在不同的中子γ混合数据集中识别中子和γ的精度、召回率和F_(1)分数均是最高,且精度、召回率和F_(1)分数能够精细化分析算法的甄别效果。The liquid scintillator detector is used to detect^(252)Cf,^(241)Am-Be and^(137)Cs.The dimensionality reduction effects of principal component analysis,kernel principal component analysis and local linear embedding in data processing are compared.K-Means clustering algorithm,Gaussian mixture model and DBSCAN clustering algorithm are used to screen the dimensionality reduction results.The deficiencies of the FOM value in some application scenarios are analyzed,and the screening effect of the algorithm is evaluated by precision,recall and F_(1)score.Comprehensive evaluation shows that the combination of nuclear principal component analysis and K-means has the best recognition effect,and the precision,recall and F_(1)score for identifying neutrons andγin different neutronγmixed data sets are the highest.And the precision,recall and F_(1)score can analyze the screening effect of the algorithm in a refined manner.

关 键 词：n/γ甄别 液体闪烁体探测器 无监督学习 聚类 FOM值 

分 类 号：O571.53[理学—粒子物理与原子核物理]