Pulse Signal-Based Sleep Classification Using Multiple Signal Processing Methods

Sleep plays a crucial role in physiological recovery, with different sleep stages exhibiting varying physiological characteristics. However, the complexity and diversity of physiological signals during sleep pose challenges to sleep scoring. Current methods for sleep stage classification are often costly and cumbersome. To address this, we propose a novel pulse signal-based approach that leverages multiple signal processing technology for accurate and portable sleep staging. We obtain pulse signals from both MIT-BIH PSG database and our self-developed pressure sensor. In our approach, we introduce a combination of feature extraction using singular spectrum analysis (SSA) and dimensionality reduction through principal component analysis (PCA). We then evaluate the effectiveness of four clustering algorithms, ultimately selecting subspace clustering, which demonstrated the best performance. Validation on 18 subjects from the MIT-BIH database and 8 volunteers yielded an accuracy above 80%, while maintaining lower hardware complexity. This demonstrates the success of our pulse signal-based method using SSA-PCA signal processing algorithms.