Research On Automatic Sleep Staging Of EEG Based On Deep Learning

Sleep quality plays an important role in the recovery and integration of the human body.Sleep staging is the basis of sleep quality assessment and a key step in the diagnosis of sleep-related diseases.At present,the medical analysis of sleep state is mainly through manual reading polysomnogram information by professional doctors to interpret the sleep stage.This process will be mixed with subjective factors,resulting in a relatively lack of objectivity in the interpretation results.This process will also consume time,so the research of automatic sleep staging model has extremely high research value and clinical application value.This thesis is researched in this context.Based on the analysis of the characteristics of EEG signals and the basic theory of sleep staging,this thesis makes a basic network of residual networks that perform well in signal classification and processing at this stage.Improved and designed a new architecture—an automatic sleep staging algorithm based on a multi-scale attention mechanism residual network and a two-way gated recurrent unit network.The main research work done is as follows:(1)In order to highlight the characteristics of EEG sleep sequence,channel feature attention unit and spatial feature attention unit are added to the residual module,and the Re LU activation function is replaced by the extended exponential linear unit activation function to construct the residual spatial channel attention module.This plays a key role in identifying different sleep periods,so that it can fully learn the importance of different channel characteristics and the correlation between characteristics.(2)In order to improve the time domain representation ability of the network for sleep sequence and solve the problem of limited learning ability of sleep cycle characteristics caused by the traditional algorithm cannot identify the timing mode in the long-term correlation data,the improved residual module is used in the same module layer to extract the feature of sleep signals with various convolution kernels of different sizes.Subsequently,the bidirectional gating loop unit network is introduced to comprehensively analyze the timing information,so as to realize the automatic learning of sleep data characteristics and the determination of sleep cycle.The experimental results show that the recognition rate of the proposed algorithm model is higher,and the biological interpretability of the model is better,which proves the feasibility and effectiveness of automatic sleep staging using residual network and bidirectional gating loop network.