TRCA-Net: using TRCA filters to boost the SSVEP classification with convolutional neural network.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCIs) are popular due to system simplicity, minimal training data needs, and high information transfer rates.
• Current SSVEP signal classification relies on knowledge-based methods like task-related component analysis (TRCA) or deep learning approaches.
• Integrating these distinct SSVEP classification methodologies has remained an unexplored area for performance enhancement.

Purpose of the Study:

• To develop and evaluate a novel algorithm, TRCA-Net, for improved SSVEP signal classification.
• To leverage the strengths of both knowledge-based TRCA and deep learning models within a unified framework.
• To enhance the signal-to-noise ratio of SSVEP data for more effective BCI communication and control.

Main Methods:

• TRCA-Net utilizes TRCA to derive spatial filters, extracting task-related components from SSVEP data.
• TRCA-filtered features are reorganized into new multi-channel signals.
• These enhanced signals are then fed into a deep convolutional neural network (CNN) for classification.

Main Results:

• TRCA-Net demonstrated significant effectiveness on two large-scale public benchmark datasets.
• Offline and online experiments with multiple subjects confirmed the robustness of the TRCA-Net approach.
• Ablation studies indicated TRCA-Net's compatibility with various CNN backbones, enhancing their performance.

Conclusions:

• The proposed TRCA-Net offers a promising advancement for SSVEP classification accuracy.
• This integrated approach has the potential to significantly improve practical applications of BCIs in communication and control.
• The study provides open-source code for the TRCA-Net algorithm, facilitating further research and development.