Automatic Sleep Staging And Features Analysis Based On EEG

In this paper, automatic sleep stage classification method based on EEG signal was studied. We classified the sleep states into four stages, awake (Wake), rapid eye movement sleep (REM), light sleep (Light), and deep sleep (Deep). Based on the fact that the frequency spectrum of EEG will change along with the transition of sleep stage, we used FFT and wavelet transformation to analyze the EEG signal in the frequency and time-frequency domains.Our database contained 20 healthy persons (33.5±14.6 years old). The sampling frequency for the EEG was 200 Hz. Besides, we also obtained three subjects from Guang An Men Hospital, China Academy of Chinese Medical Sciences (33.67±4.64 years old).Firstly, we applied the Fast Fourier transformation to the EEG signal, and then symbolized the results.Secondly, we used the symbolized results as BP neural network’s input. We used 500 sleep stages (20 subjects,10,000 sleep stages) of each subject for network training, and then we used the rest of the states (20 subjects, about 90,000 sleep stages) as a test set, to identify the four sleep states automatically.Furthermore, we conducted the six layers wavelet transformation on the EEG signal from six channels (F3-M2, F4-M1, C3-M2, C4-M1, O1-M2, O2-M1), and then got the signal of each band through inverse transformation. We analyzed the correlation between each pair of channels, and have verified the consistency between the channels.