Applying SSVEP BCI on Dynamic Background

Brain-computer interfaces (BCIs) based on steady-state visual evoked potential (SSVEP) have attracted much attention due to their high efficiency and accuracy. The SSVEP paradigm and decoding methods have been extensively studied and achieved remarkable results. This study proposed two modulation methods for the SSVEP paradigm, namely color inversion modulation and brightness compression modulation. Color inversion modulation adjusts the stimulus to adapt to the changing background, while brightness compression modulation ensures high contrast by reducing the background brightness. Furthermore, we proposed Multi-scale Temporal-Spatial Global average pooling Neural Network (MTSGNN), an end-to-end network for decoding SSVEP signals evoked by the post-modulation paradigm. MTSGNN is built with efficient convolutional structures and uses global average pooling to achieve classification, which effectively reduces the risk of model overfitting on small EEG datasets and improves classification performance. We conduct experiments to evaluate the performance of the proposed modulation and decoding methods. Compared with color inversion modulation and no modulation, the brightness compression modulation method achieved the best performance. In addition, MTSGNN outperforms the best competitive decoding method by 11.98%, 3.9% and 5.15% under color inversion modulation, brightness compression modulation and no modulation, respectively. The experimental results demonstrate the effectiveness of the proposed modulation methods and the robustness of the proposed decoding method. This study significantly improves the performance of SSVEP in dynamic backgrounds and effectively expands the practical application scenarios of BCI.