SET Net A Hybrid Deep Learning Framework For EEG Based Attention Relaxation Classification In Brain Computer...

Brain Computer Interface (BCI) technology is used as a tool to provide a direct communication channel between the human brain and the external environment. This has applications in assistive and interactive systems. The non-invasive acquisition of neural activity using Electroencephalography (EEG) is standard with BCI systems. However, it is difficult to effectively classify cognitive states, including attention and relaxation, because the EEG signal is non-stationary and noisy. To overcome these shortcomings, this study proposes a new hybrid deep learning architecture namely, squeeze-and-excitation (SE) transformer network (SET-Net). In this method, EEG signals are preprocessed and divided into temporal windows to analyze the significant spectrogram representations. The proposed model achieves a classification accuracy of 93.7%, with F1-score of 0.93 and ROC-AUC of 0.98. This demonstrates the effectiveness of the hybrid deep learning model with enhanced discrimination of EEG-based attention-relaxation state. This offers a scalable system for real-time BCI applications as assistive technology in human-computer interaction.