SSVEP Brain-Computer Interface Decoding Based on Spatio-temporal TRCA Filtering and Least Squares Data Migration

Due to its high stability and information transfer rate, steady-state visual evoked potential (SSVEP) is one of the most widely used brain-computer interface paradigms. However, because of inter-subject variability, SSVEP BCI decoding algorithms still require to be calibrated with a substantial amount of training data before practical application. This study integrates both temporal and spatial filters into the task-related component analysis algorithm to obtain more stable template signals. Additionally, transfer learning is employed to incorporate cross-subject data into the training process, reducing the number of training samples and improving the performance of SSVEP decoding algorithms. where the source domain data is migrated to the target domain using a least squares transform approach. The proposed algorithm is validated on public dataset and results show that, compared to the pure TRCA, the proposed stTRCA-LST algorithm can improve the recognition accuracy by 8.73%. The new algorithm is easy to conduct and can further simplify the calibration effort in practical application of SSVEP-based BCls.