INCEPFORMERNET: A MULTI-SCALE MULTI-HEAD ATTENTION NETWORK FOR SSVEP CLASSIFICATION

In recent years, deep learning (DL) models have demonstrated remarkable performance in Electroencephalogram (EEG) classification, especially for Steady-State Visually Evoked Potential (SSVEP)-based Brain–Computer Interface (BCI) systems. This study introduces IncepFormerNet, a novel hybrid architecture that uniquely combines Inception modules and Transformer mechanisms to enhance SSVEP classification. The Inception modules enable multi-scale temporal feature extraction by employing parallel convolutional filters of various sizes, effectively capturing subtle temporal variations in SSVEP signals. Meanwhile, the Transformer’s multi-head attention mechanism facilitates global dependency modeling, significantly improving the network’s ability to understand and represent complex signal patterns. Furthermore, filter bank techniques are incorporated to exploit the spectral characteristics of SSVEP signals. Experiments conducted on two public datasets show that IncepFormerNet achieves superior performance, reaching 87.41% accuracy on Dataset 1 and 71.97% accuracy on Dataset 2 using only a 1.0-s time window. Comparative analysis with existing DL models confirms the effectiveness and superiority of the proposed approach. The source codes in this work are available at: https://github.com/autism12138/Inceptformer .