离散小波变换结合二阶盲辨识的眼电伪迹自动去除方法

针对传统去除眼电伪迹的方法极易丢失潜在脑电信号的问题,提出一种离散小波变换(DWT)与二阶盲辨识(SOBI)结合的眼电伪迹自动去除方法(DSOBI)。首先将多通道脑电和眼电信号进行多层DWT得到多尺度下的小波系数,在小波域利用SOBI消除小波系数统计上的相关性,有效分离脑电和眼电伪迹,根据相关系数识别出眼电伪迹源分量并置零,再依次重构得到干净的脑电信号(electroencephalography,EEG)。方法对构造的数据进行去伪迹处理,均方误差为1.93,信噪比为14.32,与传统方法相比具有显著优势;对10位被试的真实脑电数据进行处理,利用相关系数验证本方法去除眼电伪迹的有效性,同时保留更多脑电信息。

Method of Ocular Artifact Automatic Removal Based on DWT and SOBI

Potential electroencephalography (EEG) signals are easily lost in traditional ocular artifact (OA) removals. A method of automatically removing ocular artifact combined with discrete wavelet transform (DWT) and second order blind identification (SOBI) was proposed, abbreviated as DSOBI. At first,obtaining the multi-scale wavelet coefficients of the collected multi-channel EEG and EOG signals through multi-layer DWT for the signals. Eliminating the statistical correlation of wavelet coefficients by SOBI in wavelet domain, separating EEG and OA effectively, OA source components were identified and removed according to correlation coefficients, and reconstructed by turns to gain the pure EEG signals. The constructed signals were processed by DSOBI with the mean squared error of 1.93 and the signal-noise ratio of 14.32, and the results compared with traditional methods showed that the performance of DSOBI is significant in OA removal. The real EEG signals from 10 subjects were processed by DSOBI and results with correlation coefficients verified that DSOBI can remove the OA effectively at the same time retain more EEG information.