Approach to recognizing SSVEP by combining CCA with artificial neural network.

Steady-state visual evoked potential (SSVEP) is one of popular EEG patterns employed by brain-computer interface (BCI) systems. SSVEP-based BCI systems have some advantages such as high information transfer rate and less training requirement for the users. To develop a SSVEP-based BCI system, the recognition of the primary frequency of SSVEP is critical and may be effectively achieved by using canonical correlation analysis (CCA). Several improved methods based on the traditional CCA method have been proposed. However, very few researches have considered the effectiveness, both by the individual subject itself and by the diversity of feeling imposed on different subjects by the same stimulus frequency, on the result of CCA. To address this issue, a novel method, NNWCCA, is proposed in this paper. NNWCCA first uses standard CCA to calculate the maximum canonical correlation coefficients between the analyzed EEG data with the reference signal of each stimulus frequency, and then employs an artificial neural network, which takes the calculated canonical correlation coefficients as inputs, to finally determine the primary frequency of the SSVEP hidden in the analyzed EEG data. The effectiveness of NNWCCA was verified by using a public SSVEP data.