摘要
目的研究一种新的脑电信号睡眠分期方法。方法利用小波包变换提取EEG信号的β和δ节律波,然后采用功率谱熵算法分别处理β和δ节律波,并整合结果得到第一部分特征参数。使用基于样本熵且尺度为11,12的多尺度熵算法分别处理EEG信号,得到第二部分特征参数。最终将所有特征参数输入到支持向量机或反向传播神经网络分类器中,将睡眠分为4期。结果对1000个睡眠脑电样本进行测试,使用支持向量机分类的平均准确率为91.90%,使用反向传播神经网络分类的平均准确率为91.70%。结论本文提出的结合小波包分解、功率谱熵和多尺度熵的方法提取的特征参数可以作为睡眠分期的有效依据,且适用于两种分类器。
Objective To propose a novel sleep staging method based on EEG signals. Methods β and δrhythms were extracted from EEG signals with wavelet packet decomposition( WPD). Then β and δ rhythms were processed with power spectral entropy( PSE) algorithm,respectively. The results of these two parts were integrated to get the first part of parameters. The multiscale entropy( MSE) algorithms based on sample entropy with scale 11,12 were used to process the EEG signals to get the second part of parameters. Finally,all the characteristic parameters were input into a support vector machine( SVM) or a back propagation neural network( BPNN) to classify sleep into four stages. Results One thousand sleep EEG samples were tested. The results showed that the accuracy using SVM was 91. 90% and the accuracy using BPNN was 91. 70%. Conclusion The proposed method combining WPD,PSE and MSE is effective for sleep staging,and it is suitable for two kinds of classifier.
出处
《航天医学与医学工程》
CAS
CSCD
北大核心
2015年第1期22-27,共6页
Space Medicine & Medical Engineering
关键词
睡眠分期
小波包分解
功率谱熵
多尺度熵
支持向量机
反向传播神经网络
sleep staging
wavelet packet decomposition
power spectral entropy
multiscale entropy
support vector machine
back propagation neural network