基于EEMD和GA-SVM的精神分裂症MEG识别  

Recognition of Schizophrenic MEG Based on EEMD and GA-SVM

作　　者：张学军[1,2] 朱丽敏[1] 黄丽亚[1,2] 成谢锋[1,2] 

机构地区：[1]南京邮电大学电子科学与工程学院,江苏南京210003 [2]江苏省射频集成与微组装工程实验室,江苏南京210003

出　　处：《计算机技术与发展》2016年第8期166-170,共5页Computer Technology and Development

基　　金：国家自然科学基金资助项目(61271334)

摘　　要：为了研究脑磁图(MEG)信号在识别精神分裂症方面的应用,文中提出了一种基于总体经验模态分解(EEMD)和希尔伯特(Hilbert)变换的识别方法。在对正常人和精神分裂症患者的脑磁信号进行预处理的基础上,通过EEMD方法将信号分解为一系列的固有模态函数(IMFs),然后使用Hilbert变换求取固有模态函数的瞬时频率和振幅,由希尔伯特-黄幅度谱(HHS)和边际谱(MS)可以发现两类信号的差异;接着选取与原信号相关性较高的前9个IMF的瞬时频率和振幅归一化后计算Hilbert加权频率;最后,利用经遗传算法(GA)优化的支持向量机(SVM)分类器进行分类,并与k近邻(KNN)分类器的结果进行对比,得到的分类精确度分别为95%和78.33%,验证了所提识别方法的有效性。In order to study the significance of magnetoencephalography （MEG） in recognition of schizophrenia,a method based on En- semble Empirical Mode Decomposition （EEMD） and Hilbert Transform （HT） is described in this paper. Firstly, on the basis of prepro- cessing of magnetoencephalography for the normal and schizophrenia, EEMD is used to decompose signals into a series of intrinsic mode functions,then instantaneous frequency and amplitude of IMFs can be obtained by HT, and the differences can be found between two types of signals from the Hilbert spectrum and the marginal spectrum. Secondly, the instantaneous frequency and amplitude of the first 9 IMFs which have higher correlation with original signals is selected to calculate the Hilbert weighted frequency after they are normalized. Finally, Support Vector Machine （SVM） optimized by Genetic Algorithm （GA） is adopted for classification, and the results are com- pared with the K-Nearest Neighbor （KNN） classifier. The accuracy of classification obtained by the two methods are 95% and 78.33% respectively, which verifies the validity of this method roughly.

关 键 词：脑磁信号 总体经验模态分解 希尔伯特变换 遗传算法 支持向量机 K近邻 

分 类 号：R318[医药卫生—生物医学工程]