Sleep staging serves as the foundation for sleep assessment and disease diagnosis, constituting a crucial aspect of sleep research. The related work on automatic sleep staging has achieved numerous satisfactory outcomes. However, current research predominantly focuses on using sleep information as classification features, e.g. employing time-domain or frequency-domain measures as local features, and comprehensive brain network information across channels as global features, while overlooking the spontaneous regularities in brain activity. Simultaneously, brain microstates are considered closely linked to brain activity and can be used to investigate the regular variations in the overall brain potential. To explore the regular changes in the microstates of brain function during sleep stages based on electroencephalogram (EEG), especially the regular changes in sleep structure, we initially conduct microstate clustering, followed by characterizing the sleep structure of the participants based on these microstates. Subsequently, we integrate the sleep structure with traditional sleep information features and perform automatic sleep staging. Our experiments make the following contributions: (1) Being the first to introduce the use of sleep structure for automatic sleep staging. (2) When there are 7 or more than 7 microstate classes, the model performs well, and the best classification accuracy reaches 89.50%. (3) Proposing a sleep automatic staging model that integrates sleep structure and sleep information.