Cross-subject spatial filter transfer method for SSVEP-EEG feature recognition.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Steady-state visual evoked potential (SSVEP) is crucial for brain-computer interface (BCI) control.
• Existing SSVEP algorithms require user-specific training data, increasing mental fatigue and limiting BCI applicability.
• Reducing interference from spontaneous electroencephalogram (EEG) activity is key to improving SSVEP recognition accuracy.

Purpose of the Study:

• To develop a cross-subject spatial filter transfer (CSSFT) method for efficient SSVEP feature decoding.
• To enable model transfer to new users without collecting their training data.
• To enhance the practical applicability of SSVEP-based BCI systems.

Main Methods:

• Proposed a cross-subject spatial filter transfer (CSSFT) method.
• Transferred existing user models with strong SSVEP responses to new user test data.
• Evaluated the method on public datasets using canonical correlation analysis (CCA) and filter bank CCA.

Main Results:

• The CSSFT method improved the distinction between target and non-target features.
• Accurate identification of incorrect targets was achieved.
• Significantly enhanced recognition performance for CCA and filter bank CCA was observed.
• Effective feature recognition was maintained even with a single data block for model calculation.

Conclusions:

• The CSSFT method eliminates the need for tedious calibration for new users.
• It offers an effective solution for cross-subject model transfer in BCI.
• The method holds significant potential for promoting wider BCI system application.