Deep Learning Method for Sleep Stages Classification by Time-Frequency Image

Classification of sleep stages is an important method in diagnosing sleep problems. This is done by experts, based on visual inspection of bio-signals such as EEG, EOGs, ECG, EMG, etc. The deep learning method is one of the newest and most important methods for analyzing, separating, and detecting images, which is becoming more and more widespread. In this paper, for the first time, the deep learning method is used to extract the EEG signal time frequency image to classify sleep stages. Here, from the one channel of EEG signal, the time frequency image of the signal is extracted and then feature extraction using the deep learning method is done. Finally, without changing the nature of the signal, the sleep steps are detected with acceptable accuracy. In this article, for the first time, time-frequency image (TFI) was provided from the one channel of the EEG signal. Then, using the AlexNet convolutional neural network by the Wigner-Ville distribution method (ANWVD), using Deeper layers contain higher-level features were extracted, and finally, using the SVM classifier, the sleep steps were classified with acceptable accuracy. The accuracy 97.6% and the time of calculations 0.36s have been reached.