Adaptive Multi-Domain Attention Network for Single-Channel Eeg Classification

Abstract

Mental attention classification using electroencephalogram (EEG) has become an important aspect of brain-computer interface (BCI) devices, especially with the growing use of single-channel EEG wearable devices. Nonetheless, the challenges of limited spatial information, vulnerability to noise, and the inability to represent features effectively are significant in the way of successful classification. Therefore, this study introduces a new Hybrid Multi-Domain Feature Fusion with Adaptive Wavelet-Entropy Attention Network (HMFF-AWEAN) to overcome these problems. The suggested framework incorporates time-domain, frequencydomain, time-frequency, and nonlinear characteristics using discrete wavelet transform (DWT), empirical mode decomposition (EMD), and entropy measurements to improve the representation of signals. In addition, a deep learning model that uses adaptive attention in combination with convolutional neural networks (CNN) and transformer architecture is suggested to identify both local and global dependencies while prioritizing important features. The experiments, performed using the Kaggle mental attention single-channel EEG dataset, showed that the proposed model has better accuracy, precision, recall, and F1-score than the baseline CNN and CNN-LSTM models. The results highlight the effectiveness of multi-domain feature extraction and adaptive attention in improving the classification performance. The suggested approach can be applied to real-time BCI applications or wearable EEG systems.