分形维数结合RLS-ICA的脑电信号消噪

Hybrid methodology combining fractal dimension and RLS-ICA for rejection of electroencephalography noise

针对脑机接口中脑电信号噪声的去除,提出将分形维数、递归式最小均方(RLS)-独立分量分析(ICA)相结合的方法.利用ICA对脑电信号进行盲源分离,得到源信号;采用分形维数自动识别源信号中的噪声信号;利用RLS自适应滤波器对已识别出来的噪声信号进行自适应滤波;通过信号重构,得到去除噪声的脑电信号.该方法有2个优点:一是通过对分形维数自动识别源信号中的噪声信号进行滤波,克服了RLS-ICA将所有源信号进行滤波,可能造成部分有用脑电信号被去除的缺点;二是通过分形维数减少RLS滤波的独立源,加快了运行速度.为了证明该方法的有效性,分别对2008年国际BCI竞赛数据和本实验室的数据进行处理.将该方法与RLS-ICA进行比较,结果显示,该方法的去噪效果明显优于RLS-ICA,单个样本的运行时间比RLS-ICA少0.07s.采用提出的方法不仅能够去除一些常见的诸如眼电(EOG)、肌电(EMG)等噪声,而且能够去除一些未知的噪声.

A novel method combining fractal dimension and recursive least-squares（RLS）-independent component analysis（ICA）was presented in order to remove noise from electroencephalography（EEG）in the study on brain computer interfaces（BCIs）.The ICA was used to decompose the contaminated EEG signals into independent components（ICs）.Then the fractal dimension was used to automatically identify ICs containing noises.The RLS adaptive filters were applied to filter noise in the identified ICs further.The processed ICs were projected back to reconstruct the uncontaminated EEG signals.The proposed method has two obvious advantages.One is that it only filters ICs identified to contain noise by fractal dimension,which can overcome the shortage that RLS-ICA filters all the ICs to result in some useful EEG being deleted.The other is that it can accelerate the speed of RLS-ICA by decreasing the number of ICs to be filtered.The 2008International BCI competition data and the laboratory data were preprocessed in order to verify the effectiveness of the proposed method.The proposed method was compared with RLS-ICA.Experimental results showed that the novel method had better performance than RLS-ICA in removing noise.The running time of one sample by the proposed method was 0.07seconds shorter than that by the RLS-ICA in average.The proposed method can not only remove electrooculogram（EOG）and electromyography（EMG）,but also remove some unknown noises.