Research On Sleep Staging Method Based On Deep Learning

Sleep is a complex process of physical activity.Sufficient sleep time and highquality sleep quality are necessary conditions for human physical and mental health.Sleep disorders seriously affect human health and life.Accurate classification of sleep stages is the key to detecting and treating sleep disorders.Sleep stage classification is also called sleep staging.In the field of sleep staging,mainstream deep learning methods only use a single relational inductive bias at the same level,which will make the feature extraction methods of deep learning methods incomplete and limit the performance of the method.This paper uses multiple relational inductive biases such as translation invariance,time invariance,and hierarchical processing.Firstly,the relational inductive bias of hierarchical processing is used to divide the sleep signal into frame level,epoch level,and sequence level according to the time domain.Secondly,by analyzing the timeinvariant nature of the sleep frame-level signal and the epoch-level signal,a multi-branch multi-scale CNN network is used to apply a translation-invariant convolutional layer to the frame-level CNN network and the epoch-level CNN network to perform sleep staging.By analyzing the timing characteristics of the sleep segment level signal,the segment level RNNs network is added to the multi-branch multi-scale CNN network to form a C/R hybrid neural network.Finally,a sequence-level RNNs network was added to learn the state transition relationship between sleep stages,and the residual connection was used to optimize the entire network.The final network is called CCRRSleep Net and is optimized using multiple non-relational inductive biases.The model proposed in this paper is tested on the Sleep-EDF dataset.In the Fpz-Cz channel without any preprocessing,an overall accuracy rate of 84.32%,and MF1 score of 79.84,and a Cohen's Kappa coefficient of 0.78 have been achieved.In the Pz-Oz channel without any preprocessing,an overall accuracy rate of 80.34%,and MF1 score of 74.63,and a Cohen's Kappa coefficient of 0.73 have been achieved.The method proposed in this paper surpasses almost all existing advanced classification methods.In summary,the CCRRSleep Net model in this paper has superior classification performance in sleep stage classification tasks,and the model construction method proposed in this paper also provides a new option for solving timing signal problems.