Multilayer correlation maximization for frequency recognition in SSVEP brain-computer interface

Abstract

Multiset canonical correlation analysis (MsetCCA) has been applied to optimize the reference signals by extracting common features in multiple sets of EEG for steady-state visual evoked potential (SSVEP) recognition. To avoid extracting the possible noise components as common features in the MsetCCA method, this study proposes an algorithm called multilayer correlation maximization (MCM) to take advantages of both the CCA method and the MsetCCA method for improving SSVEP recognition accuracy. MCM carries out three layers of correlation maximization processes: (1) correlates the original EEG data with stimulus frequency, (2) optimizes the reference signals by common features, (3) correlates the reference signals with frequency again. Experimental study is implemented to validate effectiveness of the proposed MCM method. The results indicate that the MCM method outperforms both CCA and MsetCCA for SSVEP recognition.