基于离散平稳小波和独立分量分析的事件相关电位单次提取方法

Single-trial extraction of event-related potentials based on discrete stationary wavelet and independent component analysis

如何在极低的信噪比下快速提取特征信号是事件相关电位(ERP)快速提取的关键技术。提出了将离散平稳小波和独立分量分析(ICA)相结合的方法以去除噪声,并提取事件相关电位。采用离散平稳小波变换分解ERP,选取多个尺度的子带信号,滤除高频噪声对应的小波系数;将串接小波系数作为独立分量分析的输入,利用FastICA算法实现事件相关电位的快速提取。仿真实验结果表明,与传统的相关平均法相比,该方法获得的结果较为满意;与单独采用独立分量分析方法相比,该方法的收敛速度更快。

It is very difficult to extract event-related potentials （ERP） from spontaneous rhythms under quite a low S/N ratio. Discrete stationary wavelet transformation combined with independent component analysis （ICA） was proposed to remove noises and extract event-related potential. Firstly, ERP was decomposed by discrete stationary wavelet transformation, multi-scale signals were selected. Wavelet coefficients that combined with high-frequency noises were removed. Then all wavelet coefficients were connected to perform ICA. The simulation results show that, compared with traditional averaging method, the method can simplify extraction. It also achieves a faster convergence than simple ICA method.