부산대학교 도서관

To achieve a high information transfer rate (ITR) in steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCIs), current decoding methods require extensive calibration efforts to train the model parameters for each stimulus. To facilitate the calibration process, this study proposed a cross-stimulus transfer method, which learns the common spatial filter and impulse response from a few source stimuli and then transfers them to a new target stimulus for SSVEP feature extraction. First, the common spatial filter and impulse response are obtained by minimizing the deviation between the spatially filtered source SSVEPs and the constructed SSVEP templates. Then, the feature vector comprised of two correlation coefficients is utilized for target recognition, one is the correlation coefficient between the spatially filtered target SSVEPs and the constructed templates, and the other is the canonical correlation coefficient between the spatially filtered target SSVEPs and the reference signals. For the performance evaluation, the target recognition performance of the proposed method was compared with state-of-art methods on two public SSVEP datasets and a self-collected SSVEP dataset. Results showed that the proposed method can obtain higher performance with fewer source stimuli and training blocks, demonstrating the proposed cross-stimulus transfer method has the capability of fast calibration of the SSVEP-based BCIs.