Research And Application Of Automatic Sleep Staging Algorithm For EEG Signal Based On Machine Learning

Sleep diseases are getting more and more attention from the public,and sleep state monitoring is an important assessment method for analyzing sleep quality.Traditional sleep monitoring systems obtain polysomnograms by monitoring multiple physiological signals during sleep,and then manually score them according to the existing sleep stage interpretation reference rules.This method has high requirements on the use of the site and the reader’s knowledge,and it is difficult to promote use.This thesis studies the sleep scoring model based on single channel EEG signals,and studies and designs the realization of the sleep quality monitoring auxiliary system,and strives to improve the speed and accuracy of automatic sleep staging,and reduce the difficulty of sleep quality monitoring.The main work content and research results of this thesis are as follows:(1)The combination of frequency domain filtering and eigenmode function reconstruction signal is used to reduce the interference of background noise on sleep EEG signals.First,a low-pass filter is designed to filter out power frequency and high frequency noise such as electromyography and ocular electricity.Then for some low-frequency interference and the rhythm wave of the EEG signal,there is the phenomenon of frequency band overlap.Using the adaptive filtering characteristics of empirical mode decomposition,the eigenmode function is used to reconstruct the EEG signal and reduce the interference of low-frequency artifacts.(2)Aiming at the problem of the imbalance of the number of samples in each sleep stage in the public data set,the two-phase step-by-step training method is used to reduce its impact on the performance of the classifier.In the pre-training step of the Two-phase method,random resampling and under-sampling methods are used to reconstruct the training sample set,so that the model can learn some characteristic parameters and speed up the convergence of subsequent models.In the fine-tuning step,the Focal loss function is replaced to further train the model,which effectively improves the overall performance of the automatic sleep staging model.(3)Based on the Hilbert-Huang transform theory,the time-frequency domain characteristics of sleep EEG signals are analyzed,and the first four-order eigenmode functions of EEG are used to improve the accuracy of the automatic sleep staging model.The experiment analyzed the eigenmode function of sleep EEG signal,and found that empirical mode decomposition can well highlight the rhythm wave characteristics of each period of sleep.Through statistics,it is found that the instantaneous frequency components and time-frequency energy of each eigenmode function have different distribution characteristics in each sleep stage,which provides the subsequent use of the EEG eigenmode function matrix as one of the inputs of the sleep staging model Experience knowledge support.(4)Design a CNN-LSTM network automatic sleep staging algorithm model based on single-channel EEG signals.The model uses EEG signals and eigenmode functions to form a dual-channel input structure,and its performance in predicting F1 scores in W,N1,N2,N3,and REM phases reaches 88.92%,56.79%,90.77%,82.11%,86.13%,respectively,The overall accuracy rate reached 85.6%.(5)Designed and implemented a sleep monitoring auxiliary system based on single-channel EEG.This article implements the designed automatic sleep staging algorithm into practical applications,introduces in detail the composition and implementation details of the sleep monitoring auxiliary system,and shows the application interface of the system.The system has certain practicability and easy to promote.