摘要
针对基于两种不同意识任务(想象左手运动和想象右手运动)的脑-机接口,提出采用相对小波能量的特征提取方法.首先深入研究了相对小波能量的计算方法,然后利用相对小波能量对脑电信号进行特征提取,最后采用支持向量机进行分类,并采用分类准确率和互信息作为该脑-机接口的评价标准.离线分析结果表明:分类准确率最高为85.7%,最大互信息为0.41比特.与较常用的自适应自回归(AAR)模型系数作为特征的方法相比,所提方法具有更高的识别准确率和互信息.
The feature extraction method using relative wavelet energy (RWE) is investigated for a brain-computer interface (BCI) based on two different mental tasks, i. e., the imaginary left and right hand movements. Discusses the computational method of RWE in depth, then RWE is used for the feature extraction of EEG signals with the support vector machine (SVM) used for classification. Classification accuracy and mutual information (MI) are taken as the evaluation criteria for BCI system. The off-line analysis results show that the maximum classification accuracy is 85.7% and maximum MI is 0.41 bit. Both are higher than the feature extraction characterized by the conventional adaptive autoregressive (AAR) coefficients.
出处
《东北大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
2009年第8期1103-1106,共4页
Journal of Northeastern University(Natural Science)
基金
国家自然科学基金资助项目(50435040)
关键词
脑电
脑-机接口
相对小波能量
支持向量机
互信息
EEG (electroencephalogram)
brain-computer interface
relative wavelet energy
support vector machine(SVM)
mutual information