EfficientSleepNet: A Novel Lightweight End-to-End Model for Automated Sleep Staging on Single-Channel EEG

Abstract

Sleep staging is critical for evaluating sleep quality and regulating sleep patterns. While deep learning has shown potential for automatically scoring sleep stages from raw signals, many existing models are overly complex, computationally intensive, and rely on future information, limiting their use in real-time monitoring and widespread applications. To address these challenges, we propose EfficientSleepNet, a lightweight architecture for sleep staging based on single-channel EEG. EfficientSleepNet integrates depthwise separable convolutions, grouped convolutions, channel reordering, and a novel channel attention mechanism. Compared to previous models, EfficientSleepNet significantly reduces both parameter count and complexity. We conducted extensive evaluations on two public datasets, SleepEDF-20 and SleepEDF-78, achieving accuracies of 84.4% and 80.8%, respectively, with only 83.8K parameters. These findings demonstrate that EfficientSleepNet substantially reduces model complexity and parameter count, offering strong potential for real-time sleep staging applications.