Research On Deep Learning Algorithm For Automatic Staging Of Sleep Signals

Sleep staging research has always been an important direction for sleep research.Sleep staging helps to understand sleep structure and analyze sleep quality.The current sleep problem is a very common social phenomenon.Artificial staging is not only inefficient and inaccurate,but automatic sleep staging has become the focus of sleep research.This thesis elaborates the theoretical basis of automatic sleep staging,and summarizes the research on automatic sleep staging.At the same time,different sleep signals and their characteristics,the standard and process of sleep staging and its methods are summarized and explained.In order to further study the significance of deep learning algorithms for automatic sleep staging,the research content of this thesis mainly includes two parts,the first part is the traditional machine learning-based sleep automatic staging,and the second part is the sleep automatic staging based on deep learning.Firstly,this thesis extracted 22 types of electroencephalogram signals and used support vector machine to realize automatic sleep staging.The highest accuracy is 85.93%.Combining the staging results,variance analysis has been used to further analyze the distribution of various characteristics in different sleep periods.The results show that the staging effect of non-linear features is better than time-domain analysis or time-frequency analysis,but the amount of calculation is also greater than other methods.Secondly,in order to ensure the sufficient amount of data,this thesis selected a variety of sleep signals as the original signal,including electroencephalogram signal,electrooculogram signal and electromyogram signal.Then the wavelet transform has been used to complete the preprocessing process,filtering out the mixed part of the signal and retaining its important components.A deep learning algorithm model combining onedimensional convolutional neural network and long short-term memory model has been implemented to realize automatic sleep staging.The single signal and multi-signal combinations has been discussed separately.The highest accuracy under a single signal is 93.47%,and the highest accuracy for a multi-signal combination is 93.86%.Finally,it is compared and analyzed with multilayer perceptron,convolutional neural network and long short-term memory model.The stability and accuracy of various deep learning algorithms have been discussed.The comprehensive performance and the accuracy of this algorithm is high.According to the results of this thesis,traditional machine learning sleep staging is very dependent on features.The quality of the features will affect the quality of the staging results,and the calculation of good features will be greater.The deep learning algorithm can not only avoid feature engineering,but also effectively improve accuracy,and has reference value.The deep learning algorithm model proposed in this thesis can achieve a good staging effect and has a good application prospect.