Wearable Sleep Monitoring System Based On Single-channel EEG And Deep Learning

Sleep is an important part of individual’s life cycle,and its impact on human physical and mental health and social economy cannot be ignored.As an important means of sleep research,sleep staging technology provides solutions for sleep monitoring and the diagnosis of sleep disorders.Polysomnography,known as the "gold standard" of sleep monitoring,has heavy monitoring equipment,inconvenience to carry,various measurement parameters,and many environmental constraints.Besides,it relies on sleep experts to manually classify each sleep stage,and the whole process is labor-intensive and time-consuming.Therefore,it is necessary to develop wearable automatic sleep staging technology,which has broad market prospects.This paper aims to explore the potential of sleep feature measurement and sleep representation digitization,taking single sleep EEG as the research target,starting with the design of a wearable EEG monitoring device,and using the automatic feature extraction and classification capabilities of the deep learning model to analyze the collected single-channel EEG signal,so as to achieve automatic sleep staging.The main contents of this paper are as follows:Firstly,a portable EEG acquisition device has been designed,which can realize real-time sleep monitoring at home.ADS1299,an integrated analog front-end chip,is selected to complete the realtime acquisition of sleep EEG signals.After preliminary denoising by the hardware filter circuit,24 bits Σ-Δ ADC is used to perform signal conditioning and analog-to-digital conversion on the analog signals.Then,the converted digital EEG signal is transmitted to MCU nRF52840 through SPI communication protocol for data processing.As for the packaging and transmission of EEG data,it can be directly sent to the computer or wirelessly transmit to the mobile phone by the low-power Bluetooth module to realize the real-time monitoring of sleep EEG.At the same time,it is convenient for the subsequent export of EEG data for offline processing.And the power management module is designed to supply power to each part of the system,so as to effectively reduce the power consumption of the system.Compared with the traditional polysomnography,this EEG acquisition device has the advantages of small size,light weight,low power consumption,easy to carry and low cost.It can realize long-time,continuous and high-precision EEG signal acquisition.Secondly,a single-channel EEG automatic scoring model for sleep stages based on CNNs-BiLSTM neural network has been proposed.This model optimizes the structure and parameters of Deep Sleep Net model,A which can fully exploit the feature extraction and classification capabilities of deep learning.The time-invariant features in the time domain and frequency domain are extracted from the original single-channel EEG through the dual-branch CNNs network to obtain abundant and comprehensive time-frequency information.The BiLSTM network is utilized to complete the modeling of the sleep EEG temporal relationship to fully obtain the long-term and short-term time dependencies between sleep EEG epochs,and encode the temporal information such as stage transition rules into the model.In addition,starting from the structure of CNNs-BiLSTM hybrid neural network model,a two-step training algorithm based on backpropagation has been developed,while preventing the model from suffering class imbalance problem present in a large sleep dataset.The algorithm first pre-trains the representation learning part of the model and then fine-tunes the whole model with two different learning rates.Finally,the K-fold cross validation was carried out with the single-channel EEG data in two public sleep databases MASS and Sleep-EDF.The validity of the model is also verified by horizontal comparison with the new research results in different papers in the field of automatic sleep staging.Compared with other sleep staging methods,the advantage of this model is that it can automatically learn features for sleep stage scoring from raw single-channel EEGs without using any hand-engineered features and other prior knowledge,and achieve ideal sleep staging effect.Thirdly,in order to verify the effectiveness and stability of the whole system,subjects are recruited to carry out clinical tests to continuously collect single channel sleep EEG signals.These sleep EEG signals are used as the input signals of sleep staging algorithm to obtain the results of automatic sleep staging.According to the sleep stage results predicted by the algorithm,the EEG characteristic wave and spectrum in the corresponding time period are compared and analyzed based on sleep stage standard and the characteristics of each sleep stage,so as to verify the reliability of the sleep staging results predicted by the algorithm.The whole system realizes the real-time continuous monitoring of sleep at a relatively low cost and no environmental restrictions.In addition,wearable sleep monitoring system brings more comfortable sleep experience to subjects,which has great application value for clinical sleep research and home sleep monitoring.