Sleep staging algorithm based on multi-scale combination features

In oder to solve the problem of low classification accuracy of N1 and REM in non-rapid eye movement (NREM) during sleep staging based on electroencephalogram (EEG), a sleep staging algorithm based on multi-scale combination features was proposed. This algorithm studied the influence of the characteristics of different time segments on the accuracy of N1 and REM phase segmentation. Based on the differences in waveform amplitude of N1 and REM phase and the 30s datas as the benchmark, the five-dimensional time datas were selected for feature extraction. Finally, the sample entropy, the variance and mean of wavelet coefficients, and the ratio of the absolute values of amplitudes were used to obtain the proposed features, Moreover, the support vector machine classification algorithm was used to identify the sleep period. The experimental results show that the accuracy can reach 89.43% for the time for the time segment 210s experimental data set. In addition, the average accuracy of N1 is increased by 14.99% and the accuracy of REM stage is increased by 8.20%.