基于不同神经网络算法的钚部件模板测量方法研究  

Template measurement for plutonium pit based on neural networks

作　　者：张昌繁[1] 龚建[1] 刘素萍[1] 胡广春[1] 向永春[1] 

机构地区：[1]中国工程物理研究院核物理与化学研究所,绵阳621900

出　　处：《核技术》2012年第7期525-530,共6页Nuclear Techniques

摘　　要：钚部件的模板是从申报部件的γ能谱与中子计数中提取反映其类型特征的量构成的,核查时再次测量部件的特征量并与模板进行比较判断两者是否为同一类型。将神经网络作为一种模板测量比较的匹配算法,分别应用于两种场景:BP神经网络能够对不同类别的钚部件进行分类,该场景通常用于核武库中核材料的管理与衡算,LVQ神经网络核查未知测量对象,判断是否与申报钚部件一致,该场景通常用于核裁军核查。通过实验,完善了模板的构成和匹配算法。Template measurement for plutonium pit extracts characteristic data from T-ray spectrum and the neutron counts emitted by plutonium. The characteristic data of the suspicious object are compared with data of the declared plutonium pit to verify if they are of the same type. In this paper, neural networks are enhanced as the comparison algorithm for template measurement of plutonium pit. Two kinds of neural networks are created, i.e. the BP and LVQ neural networks. They are applied in different aspects for the template measurement and identification. BP neural network is used for classification for different types of plutonium pits, which is often used for management of nuclear materials. LVQ neural network is used for comparison of inspected objects to the declared one, which is usually applied in the field of nuclear disarmament and verification.

关 键 词：钚元件 模板测量 神经网络 特征数据 

分 类 号：TJ910.89[兵器科学与技术—武器系统与运用工程] TL242.1[核科学技术—核燃料循环与材料]