A Swin Transformer-based model for monitoring attentional states with EEG signals

Accurate identification of attention states has broad applications in fields such as intelligent education, driver monitoring, and brain-computer interfaces. Electroencephalographic (EEG) signals, characterized by high temporal resolution and non-invasiveness, have become a crucial data source in attention detection research. In this study, we propose a novel attention recognition model for EEG signals, named SDNet, based on deep learning methods. The model adopts the Swin Transformer as its baseline architecture and leverages the Shifted Window Multi-head Self-Attention (SW-MSA) mechanism. A local Dense Block is introduced after SW-MSA to enhance the extraction of multi-scale spatial features. Experiments are conducted on self-collected EEG attention data. The proposed model outperforms baseline models in terms of recognition accuracy and robustness, demonstrating its effectiveness and superiority in attention state recognition tasks.