Signal extension with SeU-net for boosting the decoding performance of short-time SSVEP-based brain-computer...

Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (SSVEP-BCIs) have greatly benefited the lives of patients. However, existing SSVEP recognition methods exhibit poor performance on short SSVEP signals. SSVEP recognition accuracy heavily depends on signal length, which increases as the signal length. From a novel data perspective, this study proposes a signal extension method called SeU-net without requiring calibration data from the target subject to improve the recognition performance of calibration-free methods for short-time SSVEP signals. SeU-net employs LSTM and contrastive learning to enhance feature extraction, converting signals from sample space to feature space, and then back to the sample space to realize signal extension. SeU-net is designed to focus only on signal extension in the temporal domain, without subject-specific feature extraction operations, resulting in strong cross-subject signal extension performance. The extensive experiments demonstrate that SeU-net significantly enhances the decoding performance of calibration-free methods for short-time SSVEP signals. By enabling more accurate decoding with shorter SSVEP signals, SeU-net holds the potential to advance the practical application of high-speed SSVEP-BCIs further.