| Objective: The aim of this paper is to explore the feasibility of using this simple-to-acquire physiological(PPG)signal for sleep staging studies by using single-channel multi-featured Photoplethysmography as the study material,and to propose a sleep staging method based on the GA-DBO-RBF neural network algorithm to address the problem that traditional detection methods cannot meet the need for convenient detection at home.Methods: The study used PPG pulse wave as the research material and preprocessed the data using three methods: spline interpolation,wavelet transform,and statistical feature thresholding to eliminate baseline drift,electromyography noise,and motion artifacts.To address the issue of incomplete feature extraction for PPG pulse wave sleep staging,the study used various highdimensional feature extraction methods,such as time-domain features,frequency-domain features,and nonlinear methods,to obtain 52 features.On this basis,the study used a series of feature engineering methods(such as missing values,collinearity,important features,t-tests,and isolation forests)to extract important features.Results: By combining genetic algorithm(GA),ant colony optimization algorithm(DBO),and radial basis function(RBF)neural network,the study proposed a GA-DBO-RBF neural network algorithm-based sleep stage method and evaluated its performance.The experimental results showed that the comprehensive classification performance of the GA-DBO-RBF neural network sleep stage algorithm based on multi-feature PPG pulse wave reached 74%.Compared with using GA or DBO optimization alone,the proposed method showed better performance.Conclusion: The model proposed in this article,based on multi-feature PPG signals and using GA-DBO-RBF neural network algorithm,can be effectively used for sleep staging and achieve high classification performance.This has significant reference value for achieving more accurate and efficient home-based sleep staging detection.Moreover,the various preprocessing and feature engineering methods employed in this study also provide an implementation pathway for PPG signal processing. |
