OS-SSVEP: One-shot SSVEP classification.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Classifying steady-state visual evoked potentials (SSVEPs) is crucial for brain-computer interfaces (BCIs).
• Scarcity of calibration data, particularly with only one trial per target stimulus, presents a significant challenge for SSVEP classification accuracy.
• Existing methods struggle to effectively transfer knowledge and utilize limited target-subject data.

Purpose of the Study:

• To develop a novel approach, OS-SSVEP, for robust SSVEP classification under extreme data scarcity.
• To enhance cross-subject information transfer and leverage single-trial calibration data effectively.
• To improve the feasibility of SSVEP-based BCIs for practical applications.

Main Methods:

• Introduced OS-SSVEP, integrating a dual domain cross-subject fusion network (CSDuDoFN) with task-related and task-discriminant component analysis (TRCA and TDCA).
• CSDuDoFN employs multi-reference least-squares transformation (MLST) for domain mapping and feature fusion.
• Utilized source aliasing matrix estimation (SAME)-based data augmentation for training ensemble TRCA (eTRCA) and TDCA models.

Main Results:

• OS-SSVEP achieved state-of-the-art performance on two out of three public SSVEP datasets.
• The proposed method demonstrated competitive results on the third dataset.
• Combining OS-SSVEP with the current state-of-the-art method significantly improved classification performance.

Conclusions:

• OS-SSVEP offers a promising solution for SSVEP classification with minimal calibration data.
• The approach enhances transfer learning and data augmentation strategies for BCIs.
• This work advances the integration of SSVEP-based BCIs into daily life.