 The World Health Organization has observed an increase in the number of people suffering from somnipathy, which can cause various psychiatric and neurological disorders. Deep learning methods have been employed to develop models for sleep stage classification, but they have yet to capture the intrinsic characteristics of salient wave patterns in different sleep stages. Additionally, the class imbalance problem in the dataset makes it difficult to build a robust classification model. We propose a deep neural network that combines multi-scale extraction, MSC, and convolutional block attention module, CBM. This network extracts multi-scale features from raw EG signals and uses CBM to focus on salient variations while learning transition rules between successive sleep stages. Furthermore, a class adaptive weight cross entropy loss function is introduced to address the class imbalance issue. Experimental results on three public datasets demonstrate that our model achieves higher accuracy and macro F1 scores than other state-of-the-art approaches. This article was authored by Zhilu, Xixinluo, Yunhua Lu, and others. We are article.tv. Links in the description below.