基于二阶盲辨识结合小波包的脑电信号预处理

Electroencephalography Preprocessing Method Based on SOBI-WPD in Brain Computer Interface

针对脑机接口(BCI)中脑电信号(EEG)含有的伪迹,提出一种基于二阶盲辨识结合小波包分解(SOBI-WPD)的去除伪迹方法.首先将多个导联EEG采用SOBI分解成若干独立分量.然后根据眼电和工频干扰直观特征,将对应的独立分量置零.进而将剩余独立分量分别用‘haar’小波基进行6层WPD分解,取每个独立分量分解后与任务相关的子带进行逆变换,形成对应的新的独立分量.最后将这些新分量投影重构,得到去伪迹的EEG.对3组实验数据,使用SOBI-WPD、独立成分分析和SOBI 3种预处理方法,单个样本处理时间分别为61,239和47ms;相同的特征提取和分类方法下,识别正确率分别为86.7%,73.0%和79.8%.SOBI-WPD能快速有效地去除伪迹信号,为BCI中EEG的预处理奠定了基础.

For the artifact signal of electroencephalography（EEG）in brain computer interfaces（BCIs）,this paper proposed an artifact removal method based on second-order blind identification combined with wavelet packet decomposition（SOBI-WPD）.First,the multiple-channel EEG was decomposed into several independent components.Then,according to the intuitive characteristics of the ocular artifact and power frequency interference,the independent components of ocular artifact and power frequency were set to zero.After that,the rest of the independent components were respectively processed with sixth-order WPD based on ＂haar＂.Some decomposed and related motorimagery sub-bands of each independent component were reconstructed into a new independent component,thus,the corresponding new independent components were formed.Finally,these new independent components were projected and reconstructed with the mixing matrix to obtain EEG signals that artifacts were removed.Therefore,three methods including the proposed SOBI-WPD,independent component analysis（ICA）and SOBI were used to process three sets of experimental data.The processing time of single trial with these three methods were 61 ms,293ms and 47 ms respectively.Meanwhile,the recognition accuracy were 86.7%,73.0 % and 79.8%respectively withthe same feature extraction and classification.The proposed SOBI-WPD can quickly and effectively remove artifact signals,which may lay a foundation for preprocessing of EEG in BCIs.