A New CCA-Based Method for Improving SSVEP-Based BCI System Classification

To further improve the recognition performance of brain-computer interface (BCI) systems based on steady-state visual evoked potentials (SSVEP), several spatial filtering-based frequency recognition methods have been investigated. A new recognition method based on canonical correlation analysis (CCA) is proposed to meet expected identification effect. The method: a new CCA method is proposed based on standard CCA method, which can construct new optimized common reference templates and filters to classify electroencephalogram (EEG) signals. The standard CCA method only uses a single sine-cosine wave template at ideal frequency. The proposed method takes this problem well into account by combining the multi-channel EEG signals and the real EEG signal training data features to achieve a new reference template with high adaptability. Furthermore, the recognition results have indicated that this novel method achieve a higher recognition accuracy than other three popular methods which include the power spectral density estimation (PSDA), CCA and individual template canonical correlation analysis (ITCCA). The experiments manifest that suggested new method is able to deeply improve the recognition performance and exist promising potential for SSVEP-based BCI.