Sleep staged method based on 2D CNN-MEMM model

Abstract

Sleep staged is an important process to detect sleep quality and diagnose sleep disorders. The traditional sleep staged method has the disadvantage of insufficient accuracy and the problem of the upper limit of classification accuracy, so a method of automatic sleep stageding used jointly by 2D CNN and MEMM is proposed. Sampled from about 95 hours of sleep EEG test data from 4 subjects, the epoch-wish classification was first performed using 2D CNN and Subject-wish classification was used by MEMM. The main idea of this model is to use 2D CNN to automatically extract features from the original EEG signal, classify them by softmax, and then use the MEMM model to convert the sleep phase of the adjacent EEG cycle into a priori message, so as to improve the S2 classification performance, thereby improving the classification performance of 2D CNN. Experimental studies show that the overall accuracy of the model on the Sleep-EDF Database Expanded data set is 90.3%, and it is proved that the model can provide a way to evaluate sleep quality.