基于局域均值分解与典型相关分析的眼电伪迹去除方法

发布时间：2018-06-28 16:41

本文选题：脑电信号 + 眼电伪迹　；参考：《北京工业大学学报》2016年06期

【摘要】：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.
[Abstract]:In order to eliminate the serious effect of ocular artifact OA on electroencephalography (EEG), a new method of automatic removal of electroophthalmic artifacts based on local mean decomposition (local mean) and canonical correlation analysis (canonical correlation analysis) was proposed. Firstly, each EEG signal is adaptively decomposed into a series of production function PF components by which the correlation between PF components is removed and the corresponding typical variables are obtained. The correlation coefficient matrix between each EEG signal and multi-guide Eye-guided signal is calculated to realize the automatic recognition of eye trace components. The partial random variables corresponding to the eye trace components in the typical correlation variables are set to zero, while the other random variables remain unchanged. A new canonical correlation variable is obtained. Finally, the new canonical correlation variables are projected back to get the PF component after eye trace removal based on CCA inverse transformation, and the EEG signal after eye trace removal is further reconstructed. The experimental results based on BCI contest database show that the ratio LMDC method has better eye track removal effect than other commonly used methods, and has a strong adaptability to multiple experimenters and multiple eye tracks.
【作者单位】：北京工业大学电子信息与控制工程学院;
【基金】：国家自然科学基金资助项目(81471770) 北京市自然科学基金资助项目(7132021)
【分类号】：TN911.7;R338

【相似文献】