Sleep staging method based on CEEMDAN-WPD algorithm

To solve the problem of the current sleep staging model's low accuracy, this study performed sleep stages by extracting multidimensional features after decomposing nocturnal respiratory signals. The acoustic signal was first preprocessed. The pre-processed signals are decomposed by complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and wavelet packet decomposition (WPD), the Spearman correlation coefficients are calculated and appropriate thresholds are determined to retain the useful information components. The time-frequency domain and nonlinear features are extracted after reconstruction. Finally, it is fed into the classifier for sleep stage classification. A total of 7439 sleep acoustic signals were collected in the experiment. The experimental results show that the CEEMDAN-WPD algorithm proposed in this paper achieves an accuracy of 89.9% for REM/non-REM classification, and 84.3% for REM/light/deep sleep classification, effectively improving the classification accuracy compared with other methods.