SSVEP recognition using discriminative score fusion and transformation for portable brain computer interface

Steady State Visual Evoked Potentials (SSVEP) have been widely used in Brain-Computer Interface (BCI) systems due to their advantages, such as high signal-to-noise ratio, minimal user training, and stable spectrum property. In the last decade, Canonical Correlation Analysis (CCA) and Power Spectral Density Analysis (PSDA) have been utilized in combination with the commonly used methods for reliable SSVEP recognition. Although CCA and PSDA are considered as robust methods for SSVEP recognition, reliable SSVEP recognition is still under ongoing research investigation due to non-stationary and noisy electroencephalogram (EEG) dynamics, subject variability, and the physiological changes in the human body. This paper investigates an SSVEP-based wearable BCI system comprised of a wireless EEG recording device coupled with an Android tablet-based user interface toward daily and portable applications. We propose the fusion of CCA and PSDA at the feature level by partitioning their score spaces into multiple partitions following three different cases, extracting their heterogeneous yet complementary discriminative information, and combining them to generate a high dimensional fusion space. We also examine transforming the fusion score space to SSVEP class discriminative dimensions to mitigate the effect of redundancy using Support Vector Machine (SVM) discriminative scoring. Our experimental results demonstrate that our proposed fusion method improves the recognition accuracy from 63% to 90.78% utilizing the information at the target frequency partitions. The results further improved to 94.4% by augmenting the discriminative information in non-target partitions.