Improving the Performance of Individually Calibrated SSVEP Classification by Rhythmic Entrainment Source Separation

Abstract

The supervised decoding algorithms of Steady-State Visual Evoked Potentials (SSVEP) have achieved remarkable performance with sufficient training data. However, these methods have typically failed to achieve acceptable performance in single-trial training scenarios. To address this challenge, we propose a method to enhance SSVEP classification performance using less training data by employing Rhythmic Entrainment Source Separation (RESS) to construct spatial filters. We evaluate RESS alongside other state-of-the-art methods using two distinct datasets to assess their effectiveness. Our results indicate that RESS significantly outperforms other advanced algorithms when trained with a single block of calibration data. Specifically, compared to task-related component analysis, the RESS-based method improves average classification accuracy by 49.81% and 59.06% on the two datasets using 1-second EEG segments. The RESS-based method can significantly improve SSVEP classification performance with limited training data. RESS holds promise for practical applications in SSVEP-based BCIs, offering a novel solution to reduce the calibration data requirements for individually calibrated systems.

Keywords

• Brain-computer interface (BCI)
• steady-state visual evoked potential (SSVEP)
• rhythmic entrainment source separation (RESS)