The Improved Transformer Network Based on Temporal and Spatial Convolution

In brain-computer interface (BCI) systems, steadystate visual evoked potentials (SSVEP) are commonly used control signals that have recently seen significant advancements through deep learning (DL) techniques. Inspired by Chen J's application of Transformer models in SSVEP classification tasks, we propose the TS-Transformer network, which integrates spatial and temporal features while incorporating a multi-head self-attention mechanism to enhance EEG signal decoding accuracy. Experiments were conducted on two public datasets, covering multiple subjects to comprehensively evaluate model performance. Results show that our method achieves an accuracy of 83.77 % on the Benchmark dataset and 65.45 % on the BETA dataset with a 1second data length.