Deep Sleep Recognition Based on CNNs and Data Augmentation

Deep sleep is a key part of the sleep cycle and plays a crucial role in the daily physical recovery process. Due to the complexity and lengthy process of collecting sleep EEG data in real life, which can also pose unnecessary inconvenience to subjects over long periods, we employed three data augmentation methods to enrich the dataset and introduce more variability under limited data capabilities. In this paper, we explore the application of data augmentation in classifying deep sleep stages by analyzing electroencephalogram (EEG) signals with Convolutional Neural Networks (CNNs). These strategies are designed to present the machine learning models with a broader range of sleep EEG signal features, thereby enhancing their ability to accurately identify deep sleep stages. The study employed three publicly available CNN models that are effective for sleep stage classification and utilizes the Sleep-EDF public dataset for validation. The research findings indicate that datasets augmented with the proposed techniques show higher classification accuracy across all models, confirming the effectiveness and potential of these data augmentation methods in the context of deep sleep stage identification.