基于BP神经网络的核素快速识别算法  

Research on the Fast Recognition Algorithm of Nuclide Based on BP Neural Network

作　　者：祝美英 袁成前 刘艳芳[1] 张志鹏[1] 何琳[1] ZHU Mei-ying;YUAN Cheng-qian;LIU Yan-fang;ZHANG Zhi-peng;HE Lin(Nuclear Power Institute of China,Chengdu 610005,China)

机构地区：[1]中国核动力研究设计院,成都610005

出　　处：《核电子学与探测技术》2018年第2期284-288,共5页Nuclear Electronics & Detection Technology

摘　　要：研究了基于NaI便携式γ谱仪的核素识别算法,包括能谱光滑去噪、峰位识别、峰边界确定、基于特征峰的核素定性识别。将BP神经网络引入核素识别当中,分别采用全谱识别法和特征向量识别法对核素进行识别。全谱识别法以γ能谱每道计数作为神经网络的输入值,能充分利用能谱数据,提高结果的准确性。特征向量识别法是对能谱进行小波包分解得到频域的若干个特征值,作为神经网络的样本,大大降低了输入维数,提高了训练速度。The nuclide recognition algorithm based on NaI portable gamma spectrometer was studied,including spectral data smoothing,peak position identification,peak boundary determination,nuclide qualitative identification based on the characteristic peak matching.The BP neural network was introduced into the identification of nuclide,the full-spectrum identification method and the eigenvector identification method were used to identify the radionuclides.The full spectrum identification method uses each count of γ energy spectrum as the input value of neural network,which can make full use of energy spectrum data to improve the accuracy of the result.The eigenvector recognition method uses wavelet packet decomposition of energy spectrum to obtain several eigenvalues in the frequency domain as samples of the neural network,which greatly reduces the input dimension and improves the training speed.

关 键 词：核素识别 BP神经网络 全谱识别 特征提取 

分 类 号：TL816.2[核科学技术—核技术及应用] TP183[自动化与计算机技术—控制理论与控制工程]