Sleep Stage Classification based on BCG using Improved Deep Convolutional Generative Adversarial Networks

Monitoring sleep quality and status is important to learn health condition for improvement and prevent sleep apnea. Sleep stage classification based on Ballistocardiography (BCG) has attracted more attention due to its simplicity in equipment usage and high positioning accuracy. Firstly, this paper tries to reconstruct Electrocardiogram (ECG) signals from BCG signals using an improved Deep Convolutional Generative Adversarial Networks (DCGAN) model. Then an optimal Support Vector Machine (SVM) model is exploited for sleep stage classification, in which a Particle Swarm Optimization (PSO) is combined with a Genetic Algorithm (GA) to train the classifier. Finally, the accuracy of four-stage and six-stage SVM models are analyzed and compared using Heart Rate Variability (HRV), incorporating HRV and Respiratory Variability (RV) features. The results show that the accuracy of the four-stage and six-stage SVM models using RV features for sleep stage classification reaches 76.38% and 71.11%, respectively.