 This paper proposes a novel deep learning model, an AMRT net, to classify sleep stages from EG signals. The model uses a modified ResNet network to extract features from sub-epics of individual epics, a lightweight attention mechanism normalization based attention module, NAM, to suppress insignificant features, and a temporal convolutional network, TCN, network to capture dependencies between features of long-time series. The proposed model outperformed other state-of-the-art techniques in terms of accuracy, speed, and efficiency. This article was authored by Shuabin Su, Chen Chen, Ken Meng, and others.