Research On Signal Decomposition Algorithm And Its Application In Feature Extraction

Sleep disorders and heart diseases have brought great troubles to people,and healthy sleep and heart rhythm can promote the development and recovery of body functions.The precise classification of sleep stages and arrhythmia can provide effective references for doctors to evaluate and diagnose related diseases.Therefore,this article provides automatic detection systems for EEG signals,ECG signals,and respiratory signals,respectively.This article first preprocesses,decomposition,and extracts features of EEG,ECG,and respiratory signals.Then,extreme gradient lifting algorithm and closed-form continuous-time neural network classification method are used to achieve automatic sleep staging for EEG and respiratory signals,and fully connected neural network is used to achieve automatic classification of arrhythmia for ECG signals.This article studies three signal decomposition algorithms,including null space pursuit algorithm,cyclic singular spectrum decomposition algorithm,and empirical mode decomposition algorithm.These three algorithms are applied to the decomposition of EEG signals,ECG signals,and respiratory signals,respectively.Compared with no decomposition,the classification accuracy is improved by approximately 5%.The experimental results show that the average accuracy of using extreme gradient enhancement algorithm to stage EEG signals into 4 and 5 categories of sleep is as high as 91.07%,and the average accuracy of using closed-form continuous-time neural network to stage respiratory signals into 4 and 5 categories of sleep is as high as 94.24%.The main reason for the higher accuracy of sleep classification using respiratory signals compared to using EEG signals is the different classification methods.The extreme gradient enhancement algorithm classifies 30 second EEG signals without considering the temporal correlation of sleep,but the closed loop neural network classifies overnight respiratory signals,fully utilizing the periodicity and temporal correlation of sleep.The average accuracy of using fully connected neural networks to classify class 2 and class 4 arrhythmia in electrocardiogram signals is as high as 96.5%.The experimental results indicate that the automatic sleep staging system and arrhythmia detection system proposed in this paper using signal decomposition are reliable and feasible methods,with good application prospects.