A0790
Title: Research on sleep staging based on hidden Markov model
Authors: Bo Li - Communication University of China (China) [presenting]
Xinghui Xiao - Beijing Electronic Technology Vocational College (Austria)
ying Wang - Communication Universty of China (China)
Abstract: Sleep is an important physiological activity, and sleep staging can effectively evaluate and judge sleep structure. The purpose is to explore the implementation method of sleep staging, using 16 male subjects from the MIT-BIH multi-channel sleep database as the research object. By utilizing the temporal correlation characteristics of sleep state changes, information was extracted from electrocardiogram signals, respiratory rate, and body movement signals. A two-stage sleep staging experiment was conducted. In the first stage of sleep staging, the RR interval sequence of electrocardiogram signals was used as the observation variable, and 30 temporal, frequency domain and nonlinear features were extracted. According to the five classification results in the multi-channel sleep database, a continuous hidden Markov model was used to classify a complete sleep process. The average accuracy of the model on the test set was 64.73\%. In the second stage of sleep staging, respiratory rate and body movement signals were introduced as observation variables and first-order and second-order high-order multivariate hidden Markov models were established for classification. The average recognition accuracy was 86.29\%, and the high-order models had high consistency with the original annotations in distinguishing sleep and wakefulness, especially in correctly identifying sleep stages 1-3.
