NAMRTNet: Automatic Classification of Sleep Stages Based on Improved ResNet-TCN Network and Attention Mechanism

Sleep, as the basis for regular body functioning, can affect human health. Poor sleep conditions lead to various physical ailments, such poor immunity, memory loss, slow cognitive development, and cardiovascular diseases. Along increasing stress in society comes with a growing surge associated disorders. Studies have shown that stages are essential body’s memory, immune system, brain functioning. Therefore, automatic stage classification is of great medical practice importance monitoring conditions. Although previous research into has been promising, several challenges remain be addressed: (1) The EEG signal non-smooth harrowing feature extraction high requirements model accuracy. (2) Some existing network models suffer from overfitting gradient descent. (3) Correlation between long time sequences challenging capture. This paper proposes NAMRTNet, deep architecture based on original single-channel address these challenges. uses modified ResNet extract features sub-epochs individual epochs, lightweight attention mechanism normalization-based module (NAM) suppress insignificant features, temporal convolutional (TCN) capture dependencies series. recognition rate 20-fold cross-validation NAMRTNet Fpz-cz channel data public dataset Sleep-EDF was 86.2%. experimental results demonstrate network’s superiority this paper, surpassing some state-of-the-art techniques different evaluation metrics. Furthermore, total train 5.1 h, which much less than training other models.