 We have developed a novel multi-view CNN model to improve the accuracy of sleep staging classification. The model incorporates a margin-aware factor that takes into account the relative sample sizes of both frequent and minority classes to reduce overfitting of minority classes. This factor increases the regularization strength of minority classes without changing their sample size, thereby maximizing the decision margins of minority classes. Additionally, we have introduced two new losses, margin-aware complement entropy and margin-aware cross entropy, to further improve the performance of the model. These losses can be used together or separately to achieve different goals. Our experiments show that our proposed algorithm outperforms existing methods on three public sleep datasets, demonstrating its effectiveness in improving the accuracy of sleep staging classification. This article was authored by Fahue Miao, Li Yao, and Sio Jiexiao. We are article.tv. Links in the description below.