Sleep stage classification based on EEG is a crucial tool for understanding sleep quality. It enables the identification of various stages of sleep, including REM and non-REM sleep, which exhibit distinct patterns of brain activity and physiological responses. This valuable information can be utilized to diagnose sleep disorders, monitor treatments' impact, and better understand the correlation between sleep and overall health. Over the years, various machine learning models have been proposed to help sleep specialists identify and distinguish different stages of sleep. However, most of these models have relied on multiple EEG signals or additional modalities such as EOG or ECG, which is impractical for real-time sleep stage classification at home or home monitoring. To address this issue, we propose a novel model that leverages the transformer network to classify sleep stages based on a single EEG channel automatically. In addition, we study the impact of combining the input EEG signal with its time-frequency representations on classification accuracy. Our proposed model achieves competitive results with currently available work on highly imbalanced data sets, namely edf-20.