基于子空间分解的脑电信号眼电伪迹自动去除方法研究

摘要

To solve the electrooculogram (EOG) artifact elimination problem of electroencephalogram (EEG) signals,an automatic EOG artifact removal method is proposed based on geometric subspace decomposition.Geometric subspace is constructed by maximum noise fraction (MNF) approach,the decomposition in geometric subspace is used to separate multichannel EEG into a series of MNF components.The MNF components have the high correlation degree contain EOG artifact based on Spearman rank correlation coefficient.Thus EOG artifacts can be extracted in this detail way.Then the artifact free EEG signals are obtained by accumulating and reconstructing the processed components after projected back in signal space.Both generated data set and raw recordings were studied,combining the advantage of energy distribution visualization in brain mapping plots,the experiment results showed the effectiveness of the proposed method in artifact elimination.%针对脑电信号中的眼电伪迹去除问题,提出了一种基于几何子空间分解的眼电伪迹去除方法.最大噪声分量分析帮助构建几何子空间并将多维脑电信号分解成一系列分量,利用眼电分量间的高相关度,使用Spearman秩相关准则确定相关程度从细节中实现眼电伪迹分量的抽取;将处理后各个分量投影回信号空间并进行重构,于是在无需记录眼电的情况下得到去除眼电伪迹后的脑电信号.为了验证该方法的有效性,分别对自行叠加眼电伪迹的脑电信号及实际测量的脑电信号进行了研究,结合脑地形图能量分布可视化的优势,结果表明该方法能够对脑电信号进行有效降噪.