A Time-Local Transformation Steady-State Visual Evoked Potentials Recognition Framework Combined With Filter Banks

In steady-state visual evoked potentials (SSVEPs) decoding algorithms, applying filter banks techniques and incorporating time-local information have been proven to significantly enhance SSVEP recognition performance. However, the challenge of setting time-local parameters for each subband within the filter banks has hindered the effective integration of these two techniques, thereby limiting improvements in SSVEP algorithm performance. To address this issue, we propose an SSVEP recognition framework that combines filter banks and time-local transformation (CFBTT). This framework is based on our hypothesis that increasing the energy distribution of SSVEP components within each subband can enhance algorithm performance. Therefore, we determine the time-local information parameters for each subband based on the differences in energy distribution before and after the time-local transformation and develop two parameter optimization methods designed for training-free and training algorithms, {M}1 and {M}2 , effectively integrating filter banks techniques with time-local transformation. Experimental results on two large-scale SSVEP datasets demonstrate that applying the CFBTT framework leads to significant improvements in recognition performance for both training-free and training algorithms. The enhanced performance of multiple state-of-the-art algorithms validates our hypothesis and demonstrates that our framework possesses strong generalization capabilities and broad applicability.