经验模式分解与代价敏感支持向量机在癫痫脑电信号分类中的应用

摘要

EEG signals can be extracted from EEG signals, which can better understand the characteristics of EEG signals.However, due to the aliasing of various types of external signals, the signal exhibits nonlinear and nonstationarity.Therefore, for EEG signals, extraction is a problem.In this paper, an empirical mode decomposition (EMD) algorithm, which is superior to wavelet decomposition, was proposed to decompose the EEG signal and extract the eigenvalues of the main IMF components.Then, the cost-sensitive support vector machine (CSVM) was used to classify the parameters excellent.In the study of EEG signals of epilepsy patients, the accuracy of classification is more than 90%, which verifies the feasibility of this method.%提取出脑电信号中微弱征兆信息,可以更好地了解脑电信号的特征,但由于各类外界信号的相互混叠,信号呈现出非线性、非平稳性,因此脑电信号的提取是个难题.为此本研究提出了优于小波分解的经验模式分解(EMD)算法对脑电信号进行分解,提取主要IMF分量的特征值,随后采取代价敏感支持向量机(CSVM)进行分类,并对参数进行寻优.在对癫痫患者脑电信号研究的实验中,分类准确率均达到90%以上,验证了本方法的可行性.