Fixed template network and dynamic template network: novel network designs for decoding steady-state visual evoked...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Steady-state visual evoked potentials (SSVEPs) are crucial for brain-computer interfaces (BCIs).
• Decomposition methods are efficient for SSVEP decoding, but deep learning (DL) models are emerging.
• Current DL models for SSVEP decoding lack fair comparisons with state-of-the-art decomposition methods.

Purpose of the Study:

• To propose novel deep learning network designs for SSVEP decoding.
• To compare the performance of these novel networks against established decomposition methods.
• To introduce a data augmentation technique for SSVEP datasets.

Main Methods:

• Introduced Fixed Template Network (FTN) and Dynamic Template Network (DTN), combining fixed and subject-specific template advantages.
• Developed a novel data augmentation method tailored for SSVEP signals.
• Conducted intra-subject classification performance comparisons on three public SSVEP datasets.

Main Results:

• Both FTN and DTN demonstrated suboptimal classification performance compared to state-of-the-art decomposition methods.
• The proposed networks achieved competitive results within the SSVEP decoding landscape.
• The data augmentation method showed potential for improving SSVEP signal processing.

Conclusions:

• The novel FTN and DTN network designs show potential for enhancing SSVEP decoding performance.
• These networks offer a promising direction for improving the practicality of SSVEP-based BCIs.
• Further research is warranted to optimize these deep learning approaches for SSVEP applications.