Mixup-based data augmentation for enhancing few-shot SSVEP detection performance.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Few-shot steady-state visual evoked potential (SSVEP) detection is crucial for brain-computer interfaces (BCIs).
• Limited calibration data in SSVEP detection often compromises BCI system performance.
• Developing effective data augmentation (DA) strategies is essential for improving few-shot SSVEP detection.

Purpose of the Study:

• To enhance few-shot SSVEP detection performance in BCI systems.
• To introduce and validate a novel mixup-based data augmentation strategy for SSVEP signals.
• To reduce the calibration time and improve the practical usability of BCI systems.

Main Methods:

• A mixup-based data augmentation method was proposed, generating synthetic SSVEP trials via linear interpolation.
• The interpolation weight was optimized by maximizing signal similarity to template and reference signals.
• Augmented data was utilized to train spatial filters, employing task-related component analysis with neighboring stimuli data.

Main Results:

• The mixup-based DA method significantly improved SSVEP detection accuracy under few-shot conditions.
• The proposed method outperformed existing data augmentation and baseline approaches on two benchmark datasets.
• Enhanced performance was observed in SSVEP decoding with limited available data.

Conclusions:

• The mixup-based DA method provides an effective solution for improving SSVEP decoding with minimal data.
• This approach can reduce calibration requirements, thereby enhancing the real-world applicability of BCI systems.
• The strategy offers a practical means to boost the overall usability and performance of BCI technology.