Research On EEG Signal Sleep Staging Method Based On Deep Learning

With the development of technology and the increasing pace of life,sleep problems have become a serious problem in people’s daily lives and it is vital to help people assess their sleep quality to improve their work efficiency and quality of life.Sleep staging is a prerequisite for sleep quality assessment,and it can be used to determine the sleep quality of patients by understanding the distribution of their sleep states.In this paper,deep learning methods are applied to sleep EEG signal staging,from sleep EEG signal feature characterization,sleep EEG signal data enhancement,and automatic sleep staging model improvement.The main work of this thesis is as follows:In response to the inefficiency of manual methods for sleep stage classification,this paper introduces Convolutional Neural Network(CNN)into sleep staging and conducts research on the way to characterize sleep EEG signal features input to the classification network by combining the characteristics of sleep EEG signals.Based on the advantages of CNN for picture feature extraction,the Short time Fourier Transform(STFT),Continuous Wavelet Transform(CWT)and Hilbert-Huang Transform(HHT)are selected to be used for the classification of one-dimensional EEG signals.The STFT-CNN,CWT-CNN and HHT-CNN classification models were constructed using three analysis methods that can transform one-dimensional EEG signals into two-dimensional time-frequency spectrograms.The results show that wavelet time-frequency spectrograms are the most effective in the automatic sleep staging study.To address the problem of unbalanced EEG data sets in different sleep stages leading to poor sleep staging,this paper proposes an RDB-DCGAN data enhancement model with L loss function based on Deep Convolutional Generative Adversarial Network(DCGAN).The model takes two-dimensional continuous wavelet time-frequency maps as input,augments a few classes of sleep EEG data,and then performs sleep staging through CNN networks.The results of CNN classification tests on the publicly available Sleep-EDF dataset showed that the accuracy of sleep staging at all stages improved after data augmentation,especially at the N1 stage where the classification accuracy was low due to the small amount of original data,demonstrating that data augmentation with the improved DCGAN model can effectively improve the classification of class-imbalanced sleep datasets.To address the problem that a single CNN network cannot fully extract the deep features of the input time-frequency map,this paper combines the multi-scale feature extraction module and the SE attention mechanism to build a multi-scale automatic sleep model with shallow feature extraction,multi-scale feature extraction and deep feature extraction on the basis of CNN networks.Staging model.The results of the Sleep-EDF dataset showed that the improved automatic sleep staging model can effectively improve the accuracy of sleep staging and achieve more accurate sleep staging.