An improved cross-subject spatial filter transfer method for SSVEP-based BCI.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Steady-state visual evoked potential (SSVEP) recognition algorithms require user training data, which can be tedious and cause mental fatigue.
• This limits the practicality of SSVEP-based brain-computer interface (BCI) systems.
• Existing cross-subject spatial filter transfer (CSSFT) methods use superposition averages but struggle with trial-to-trial signal variations.

Purpose of the Study:

• To improve the superposition averaging technique within the CSSFT method.
• To propose novel Ensemble and Expansion schemes for enhanced SSVEP feature recognition.
• To reduce the need for extensive user-specific training data in BCI applications.

Main Methods:

• Development of an improved CSSFT method incorporating Ensemble and Expansion schemes.
• Utilizing superposition averaging with enhanced techniques for data transfer.
• Comparative analysis of the proposed method against the standard CSSFT using public datasets.

Main Results:

• The improved CSSFT method demonstrated significantly enhanced feature recognition performance.
• Key metrics such as recognition accuracy and information transmission rate were notably improved.
• The proposed schemes effectively addressed variations in brain signal amplitude and patterns across trials.

Conclusions:

• The developed strategy avoids tedious data collection processes for SSVEP-based BCI systems.
• This approach significantly improves recognition accuracy and information transmission rates.
• The findings promote the practical application of SSVEP-based BCI technology by increasing efficiency and user-friendliness.