基于GASF与MSVM的放射性核素识别方法  

Radionuclide Recognition Method Based on GASF and MSVM

作　　者：周思益 张江梅[1,2] 刘灏霖 冯兴华 张草林 ZHOU Siyi;ZHANG Jiangmei;LIU Haolin;FENG Xinghua;ZHANG Caolin(School of Information Engineering,Southwest University of Science and Technology,Mianyang 621010,Sichuan,China;Fundamental Science on Nuclear Wastes and Environmental Safety,Southwest University of Science and Technology,Mianyang 621010,Sichuan,China;Department of Automation,University of Science and Technology of China,Hefei 230026,Anhui,China)

机构地区：[1]西南科技大学信息工程学院,四川绵阳621010 [2]核废物与环境安全国防重点学科实验室,四川绵阳621010 [3]中国科学技术大学自动化系,合肥230026

出　　处：《西南科技大学学报》2023年第2期78-84,共7页Journal of Southwest University of Science and Technology

基　　金：国防基础科研计划(JCKY2020404C004)。

摘　　要：提出了一种基于格拉姆角和场(GASF)与基于Mahalanobis距离的支持向量机(MSVM)的核素识别方法。将核素γ能谱数据视为一维序列,利用GASF方法将能谱数据二维化,再利用双向二维主成分分析对二维化能谱数据进行降维以进行特征提取,设计MSVM分类器并结合遗传算法进行参数寻优,实现对γ能谱(核素)的识别,利用Geant 4仿真核素γ能谱数据对本文算法与寻峰算法、SVD-SVM算法进行了对比实验,同时在真实核素γ能谱数据上进行了识别实验。结果表明:本文方法与同类方法相比,通过利用全谱信息,有效提高了核素识别准确率;在探测距离为20 cm内,对真实探测环境中得到的不同探测距离的核素能谱的平均识别率均高于96%,表现出良好的识别性能。A nuclide recognition method based on Gram angle sum field(GASF)and support vector machine based on Mahalanobis distance(MSVM)was proposed.The nuclideγenergy spectrum data was regarded as a one-dimensional sequence,the GASF method was used to make the two-dimensional energy spectrum data,and the bidirectional two-dimensional principal component analysis was used to reduce the dimensionality of the two-dimensional energy spectrum data for feature extraction.The MSVM classifier was designed and the parameter optimization was conducted combined with genetic algorithm to realize the recognition ofγenergy spectrum(nuclide).The Geant 4 simulation nuclideγspectrum data was used to compare the proposed algorithm with the peak search algorithm and the SVD-SVM algorithm,and the identification experiment was carried out on the real nuclideγspectrum data.The results show that compared with other methods,the proposed method can effectively improve the accuracy of nuclide identification by using the full spectrum information.Within a detection distance of 20 cm,the average recognition rate of the nuclide energy spectra obtained at different detection distances in real detection environment is higher than 96%,showing good recognition performance.

关 键 词：Γ能谱 核素识别技术 格拉姆角和场 双向二维主成分分析 MAHALANOBIS距离 支持向量机 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]