Recognition of Frequencies of Short-Time SSVEP Signals Utilizing an SSCCA-Based Spatio-Spectral Feature Fusion Framework

Abstract

A brain-computer interface (BCI) facilitates direct communication between the brain and external equipment through EEG, which is preferred for its superior temporal resolution. Among EEG techniques, the steady-state visual evoked potential (SSVEP) is favored due to its robust signal-to-noise ratio, minimal training demands, and elevated information transmission rate. Frequency detection in SSVEP-based brain-computer interfaces commonly employs canonical correlation analysis (CCA). SSCCA (spatio-spectral canonical correlation analysis) augments CCA by refining spatial filtering. This paper presents a multistage feature fusion methodology for short-duration SSVEP frequency identification, employing SSCCA with template signals derived via leave-one-out cross-validation (LOOCV). A filterbank generates bandpass filters for stimulus frequencies and their harmonics, whereas SSCCA calculates correlation coefficients between subbands and templates. Two phases of non-linear weighting amalgamate these coefficients to discern the target stimulus. This multistage methodology surpasses traditional techniques, attaining a accuracy of 94.5%.