Cross-Subject Transfer Learning for Boosting Recognition Performance in SSVEP-Based BCIs.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCIs) offer high communication rates and signal quality.
• Transfer learning is crucial for enhancing SSVEP-BCI performance by leveraging auxiliary data.
• Existing methods often require significant subject-specific training data.

Purpose of the Study:

• To propose an inter-subject transfer learning method for improving SSVEP recognition.
• To enhance SSVEP detection accuracy by utilizing transferred templates and spatial filters.
• To reduce the reliance on extensive subject-specific training data.

Main Methods:

• A novel inter-subject transfer learning framework was developed.
• Spatial filters were trained using multiple covariance maximization to extract SSVEP information.
• Transferred templates and spatial filters were generated using regression and covariance maximization.
• A four-dimensional feature vector was constructed for SSVEP detection.

Main Results:

• The proposed method demonstrated improved SSVEP recognition performance.
• Validation was conducted using both a public and a self-collected dataset.
• Experimental results confirmed the feasibility and effectiveness of the inter-subject transfer learning approach.
• Contribution scores were calculated to quantify source subject influence.

Conclusions:

• The developed inter-subject transfer learning method effectively enhances SSVEP detection in BCIs.
• The approach shows promise for improving BCI performance across different subjects.
• This method offers a viable solution for reducing subject-specific training requirements in SSVEP-BCIs.