单通道脑电信号中眼电干扰的自动分离方法

当前主流的眼电(EOG)去除方法需要利用多通道脑电的相关性，难以在单通道的便携式脑机接口(BCI)中应用。该文提出一种基于长时差分振幅包络与小波变换的眼电干扰自动分离方法。首先在原脑电信号的长时差分振幅包络上实施双门限法来精确检测眼电的起止点，然后利用sym5小波对脑电进行分解并引进Birgé_Massart策略来自适应地确定小波重构系数阈值，最后通过小波重构精确地估计眼电，实现单通道上眼电与脑电的自动分离。大量实验证明，该方法与主流的平均伪迹回归分析和基于独立成分分析(ICA)的方法相比，能够获得更好的估计眼电与原眼电的相关性，保证更高的校正信噪比和较强的实时性，能够满足脑机接口多方面的需要。

Automatic Electrooculogram Separation Method for Single Channel Electroencephalogram Signals

The traditional ElectroOculoGram (EOG) correction methods usually use the correlation information of multi-channel ElectroEncephaloGram (EEG), and are difficult to apply to portable Brain-Computer Interface (BCI) in single channel. An automatic EOG separation method is proposed based on the long term difference amplitude envelope and the wavelet transformation in the paper. Firstly, the accurate EOG beginning and ending points are detected on the long term difference amplitude envelope of the original EEG through a dual thresholds method. Secondly, the sym5 wavelet is applied to decompose the original EEG signal, and the Birgé_Massart strategy is introduced to adaptively determine the thresholds of wavelet coefficients. Finally, the EOG is accurately reconstructed and separated from the EEG in this channel. Compared with the popular regression analysis of averaging artifact and the Independent Component Analysis (ICA) based methods, the proposed method is proved to achieve a better correlation measure between the separated EOG and the original EOG, a higher signal-to-noise ratio of the corrected EEG, and a good real-time operating speed for most BCI application requirements.