Research On Automatic Sleep Staging Method Based On Deep Convolutional Neural Network

About a third of a person's life is spent in sleep.Sleep is an essential part of the normal functioning of the immune system,which is essential for everyone.Good sleep can promote people's growth and development,alleviate fatigue and so on,is the basis of maintaining life.However,with the increasing pressure of social life,the influence of sleep-related diseases on human beings is also increasing,which has promoted the development of sleep medicine research.The research of sleep is primarily concerned with the study of sleep stages.The use of Electroencephalograph(EEG)recordings,which reflect a range of brain activity,to study sleep is important in assessing sleep stage and quality.In recent years,the combination of signal processing and deep learning for sleep staging has gradually become the mainstream.However,there is still a big gap between China and foreign countries in terms of the research progress of sleep stages and the accuracy of sleep stages.In this paper,a new method for automatic sleep staging is proposed.Firstly,a Convolutional Neural network(CNN)is constructed,then the network is used to extract the characteristics,and finally the sparse representation-based classifier(SRC)is combined to classify the sleep.This method first uses the wavelet transform for time-frequency processing to the original EEG signals,the GoogLe Net as a feature extractor from the resulting frequency according to its inherent in mining the deep brain electrical characteristics,and puts forward will be replaced with the new GoogLe Net last four layer network layer to achieve the purpose of more accurate classification,so as to realize the automatic stage of sleep convolution neural network construction.Finally,the fully connected layer of the newly constructed network is discarded,and the SRC is used to replace the fully connected layer.The image features that should be fed into the fully connected layer are fed into the SRC,and the sleep stages are made by combining with the SRC.The ultimate learning machine was used as a control group to test the accuracy of the proposed method.In this paper,the performance of the constructed deep convolutional neural network wasevaluated by observing the accuracy of sleep stages with the EEG data of 25 sleep volunteers.In the experiment,the average staging accuracy of the method proposed in this paper,which combines the creation network with the SRC,reached 78.68%.In terms of time,the longest time to obtain the experimental results is about 23 minutes,and the average time is about 18 minutes,which proves that the method proposed in this paper not only has good staging performance,but also can effectively shorten the experimental time and improve the staging efficiency,thus providing an effective method for sleep research.Future work will focus on how to increase the accuracy of sleep stages and improve the practicability of the method without increasing the complexity and running time of the model.