改进独立分量算法的眼电伪迹去除方法研究

脑电信号采集过程中易受眼电干扰，给脑电信号分析处理带来极大的不便，由此提出了一种改进独立分量分析（IICA）自动去除眼电伪迹的方法。该方法将水平和垂直眼电信号按照一定的比例混叠成一导新的信号，并与脑电信号一起作为输入；采用基于负熵判据的FastICA算法快速获取各导独立分量；记录此时的负熵判据参数a]，并利用相关系数识别混叠眼电信号独立分量，记录对应的相关系数；a]加上一定的步长，重复上述步骤至a]达到阈值时停止；重复多次上述循环，获取均值向量，取出均值向量中最大的相关系数与所对应的a]，根据a]获取新的独立分量，采用相关系数自动识别混叠眼电独立分量，并置零；再进行ICA逆变换返回到原信号各个电极，即可得到同时去除水平与垂直眼电伪迹后的各导脑电信号。实验结果表明，IICA方法能有效降低去伪迹耗时，极大提高信噪比，减少均方根误差。

Research on improved independent component analysis to ocular artifacts removal from EEG signals

Electroencephalogram（EEG） is easily affected by Ocular Artifacts（OA）, which would be harmful to analysis. The Improved Independent Component Analysis（IICA）is a novel method for Ocular Artifacts removing automatically. Firstly, the horizontal and vertical electro-oculogram aliasing together, and together with EEG as the input, the independent components are gained through the FastICA algorithm. Secondly, it records the negative entropy criterion parameter, uses the correlation coefficient to recognize the independent component and records the corresponding correlation coefficient. Thirdly, the parameter adds steps. Then repeats the above steps until the parameter achieves threshold. Fourthly, pick up the maximum coefficient of the above coefficients and the corresponding parameter. Finally, use the new parameter to gain the independent components, and use correlation coefficient to recognize the aliasing signal component. The EEG without Ocular Artifacts are reconstructed using inverse transformation of ICA. Experimental results show that IICA lowers time-consuming, and improves the signal-to-noise ratio, and reduces root-mean-square errors.