Nuanprasert Somchai, Adachi Yoshiaki, Suzuki Takashi
Annu Int Conf IEEE Eng Med Biol Soc. 2015 Aug;2015:2767-70. doi: 10.1109/EMBC.2015.7318965.
In this paper, we present the noise reduction method for a multichannel measurement system where the true underlying signal is spatially low-rank and contaminated by spatially correlated noise. Our proposed formulation applies generalized singular value decomposition (GSVD) with signal recovery approach to extend the conventional subspace-based methods for performing the spatio-temporal filtering. Without necessarily requiring the noise covariance data in advance, the implemented optimization scheme allows users to choose the denoising function, F(·) flexibly satisfying for different temporal noise characteristics from a variety of existing efficient temporal filters. An effectiveness of proposed method is demonstrated by yielding the better accuracy for the brain source estimation on simulated magnetoencephalography (MEG) experiments than some traditional methods, e.g., principal component analysis (PCA), robust principal component analysis (RPCA) and multivariate wavelet denoising (MWD).
在本文中，我们提出了一种用于多通道测量系统的降噪方法，该系统中真实的潜在信号在空间上是低秩的，并受到空间相关噪声的污染。我们提出的公式应用广义奇异值分解（GSVD）和信号恢复方法来扩展传统的基于子空间的方法，以执行时空滤波。所实现的优化方案无需事先获取噪声协方差数据，允许用户灵活选择去噪函数F(·)，以满足来自各种现有高效时间滤波器的不同时间噪声特性。通过在模拟脑磁图（MEG）实验中进行脑源估计，与一些传统方法（例如主成分分析（PCA）、稳健主成分分析（RPCA）和多变量小波去噪（MWD））相比，所提方法具有更高的准确性，从而证明了该方法的有效性。
