摘要
传统的核素识别主要根据能谱中特征峰确定核材料中放射性核素的种类,当出现叠峰时,常规方法就无法实现核素识别。采用逻辑回归二分类的算法进行核素的识别,利用小波包分解将能谱拆成不同频率信号,再将不同频率信号进行重构,计算频率信号的特征信号。把这个特性信号看作能量,将能量组成特征向量,会得到一组与信号相对应的能量序列,可构成一组特征向量。对测量的所有γ能谱进行特征向量提取用作机器学习的训练集和测试集。将训练集代入进行预测函数模型训练。通过求解损失函数全局最小值得到预测函数模型最优解的参数θ。代入测试集计算核素识别正确率为97%。经过实验验证了所提算法的可行性,对快速识别核素具有一定的实际价值。
The traditional nuclide identification mainly determines the types of radionuclides in nuclear materials according to the characteristic peaks in the energy spectrum.When there are overlapping peaks,the conventional method can not realize nuclide identification.The algorithm of logistic regression binary classification is proposed to identify nuclides.The energy spectrum is divided into different frequency signals by wavelet packet decomposition,and then the different frequency signals are reconstructed to calculate the characteristic signal of frequency signal.This characteristic signal is energy,and the energy is combined into characteristic vector.A set of energy sequences corresponding to the signal will be obtained,which can form a set of eigenvectors.For all measuredγthe energy spectrum is used to extract the feature vector,which is used as the training set and test set of machine learning.The training set is substituted to train the prediction function model,and the parameters of the optimal solution of the prediction function model are obtained by solving the global minimum value of the loss functionθ.The accuracy of nuclide recognition calculated by substituting into the test set is 97%.The feasibility of the proposed algorithm is verified by experiments,which has a certain practical significance for rapid nuclide identification.
作者
周文清
周达
康建军
ZHOU Wen-qing;ZHOU Da;KANG Jian-jun(National Marine Technology Center,Tianjin 300112,China)
出处
《核电子学与探测技术》
CAS
北大核心
2023年第1期12-17,共6页
Nuclear Electronics & Detection Technology
关键词
机器学习
特征提取
逻辑回归
核素识别
损失函数
machine learming
feature extraction
logistic regression
nuclide recognition
loss function