基于快速独立分量分析和小波包变换的脑电信号消噪

Noise of EEG Eliminated by the Combination of Fast Independent Component Analysis and Wavelet Packet Transform

本文提出一种小波包变换和快速独立分量分析相结合的方法对脑电信号进行预处理。首先利用小波包变换对脑电信号进行3层分解,对第三层的高频小波系数置零,以此达到去除随机噪声的目的,同时最大程度地保留了细节信息。其次,对经过小波包变换处理后的多路信号进行快速独立分量分析,将脑电信号与各噪声源信号分离开。为验证消噪算法的效果,本文对输出各分量进行相关性分析,实验结果显示各分量间的互相关系数的数量级为-15和-16,互相关系数近似为0,说明该方法的去噪效果很好。

In this paper, the approach of the combination of Fast ICA and wavelet packet transform is proposed to process the EEG signals. First of all, the EEG signals were decomposed up to 3 levels using wavelet packet transform. Then put coefficients of the highest frequency range at zero, so as to eliminate random noise in EEG signals, and at the same time to keep as much detailed information as possible. After that, in order to separate all the noise from EEG signals, the output of the wavelet packet transform was processed by Fast ICA. To verify the performance of the system, the author analyzed the correlations among all components of the result. The result of the experiment showed that the order of magnitude of cross correlation coefficients were-15 and-16, which could be regarded as zero. So this approach performed well in reducing noise.