基于局域均值分解与典型相关分析的眼电伪迹去除方法

Ocular Artifact Removal Based on Local Mean Decomposition and Canonical Correlation Analysis

为消除眼电伪迹(ocular artifact,OA)对脑电信号(electroencephalography,EEG)造成的严重影响,提出一种基于局域均值分解法(local mean decomposition,LMD)与典型相关分析(canonical correlation analysis,CCA)的眼电伪迹自动去除方法,记为LMDC法.首先,利用LMD将每导脑电采集信号自适应地分解为一系列具有物理意义的乘积函数(production function,PF)分量,通过CCA去除PF分量之间的相关性,获得相应的典型变量;其次,计算每导脑电信号与多导眼电信号间的相关系数矩阵,实现眼迹成分的自动识别,将典型相关变量中对应眼迹成分的部分随机变量置零,其余随机变量不变,得到新的典型相关变量;最后,基于CCA逆变换将新的典型相关变量投影返回得到眼迹去除后的PF分量,并进一步重构出眼迹去除后的脑电信号.基于BCI竞赛数据库进行实验研究,结果表明:LMDC法相对其他常用方法获得了较好的眼迹去除效果,并对多位实验者和多种眼迹表现出较强的自适应性.

Based on local mean decomposition （LMD） and canonical correlation analysis （CCA）, anautomatic removal method, denoted as LMDC, was proposed to eliminate the serious impact of ocularartifact （OA） from electroencephalography （EEG）. Each recorded EEG was decomposed into a series ofphysically meaningful production function （PF） components adaptively by LMD, and CCA was applied toeliminate the correlation among the PFs to get the corresponding canonical correlation variable. Then, thecorrelation coefficient matrix between each EEG and multi electrooculogram （EOG） was computed torecognize the OA component automatically. The random variables corresponding with OA components inthe canonical correlation variable were set to zero, and the others remain unchanged to obtain a newcanonical correlation variable. Finally, the inverse algorithm of CCA was utilized to project the newcanonical correlation variable to the OA free PFs, and the OA removed EEG was reconstructed.Experimental research was conducted on a public brain computer interface （BCI） completion database.Experiment results show that LMDC has better performance than that of the other related methods, andhas stronger adaptability for multi subjects and types of OA.