Compressive Sensing of Multichannel EEG Signals Based on Graph Fourier Transform and Cosparsity

摘要

Cosparsity as a useful prior has been extensively applied in accurate compressive sensing (CS) recovery of multichannel electroencephalogram (EEG) signals from only a few measurements. Latest studies proved that exploiting cosparsity and channel correlation in a unified framework can obtain accurate recovery results. However, all these methods ignore the adjacent relationship between the real physical electrodes and exploit the inaccurate channel correlation. Another problem is that most methods employ convex regularizations to exploit cosparsity and channel correlation, which cannot obtain competitive results. In this paper, a novel graph Fourier transform and nonconvex optimization (GFTN)-based method is proposed to enforce inherent correlation across different channels and cosparsity. Alternative direction method of multipliers is used to solve the resulting nonconvex optimization problem. Experiments show that GFTN can remarkably improve the performance of CS recovery for multichannel EEG signals.