Motor imagery EEG signal classification based on improved CTNet

Brain-computer interface (BCI) technology enables direct communication between the brain and external devices, opening new possibilities for human-computer interaction and rehabilitation applications. Focusing on the critical task of motor imagery EEG (MI-EEG) signal decoding, this study proposes an enhanced CTNet model incorporating multi-scale convolutional networks and hybrid attention mechanisms (MHA-CBAM) for improved classification performance. The model's key innovations include: (1) a multi-scale temporal feature extraction module capturing EEG patterns across different time windows, and (2) an integrated attention mechanism that combines global sequence modeling with local channel feature enhancement. Comprehensive evaluations on BCI Competition IV datasets demonstrate the model's effectiveness, achieving classification accuracies of 83.87% (4-class IV-2a) and 88.79% (2-class IV-2b), representing improvements of 1.35% and 0.30% respectively over the baseline CTNet architecture. These advancements significantly advance the state-of-the-art in MI-EEG decoding for practical BCI systems.