Research On Motion Artifact Removal And Feature Extraction Algorithm Of PPG Signal

The blood pressure,as a physiological index,is vital for human health,and timely monitoring of blood pressure can effectively prevent cardiovascular and cerebrovascular diseases.With the development of signal processing of sensor technology,At present,it is more convenient to monitor blood pressure for a long time by using Photopleethysmograph(PPG)to obtain human physiological signals to realize blood pressure prediction.In the blood pressure monitoring method based on PPG,when the tester is in a non-static state,motion artifacts will be introduced into the obtained pulse wave signal,which will adversely affect the PPG signal,make the waveform lose original characteristics,and reduce the overall accuracy and effectiveness of blood pressure prediction using PPG signal.Due to the different motion artifacts in different motion states,incomplete artifact removal or excessive filtering will occur in denoising,which is not conducive to blood pressure monitoring.Aiming at the above problems,a method of artifact removal and blood pressure estimation of PPG signal is proposed.The main research contents include two aspects:artifact removal and feature extraction.1.In the stage of artifact removal,Based on the analysis of different types of motion in daily life,a joint FIR-KNN-wavelet denoising method is proposed according to the characteristics of artifacts.In this method,a band-stop filter is designed according to the acceleration information to filter out the main motion artifacts.A statistical parameter kurtosis and skewness combined with K nearest neighbor classification method is proposed to classify PPG signals after FIR bandstop filtering.The joint wavelet threshold denoising of PPG signal which needs secondary filtering is found out.Compared with the single FIR band-stop filtering method and the unclassified joint filtering method,the peak-to-peak amplitude variance and the peak-to-peak interval time variance of the waveform after removing the motion artifacts by the FIR-KNNwavelet filtering method are increased by 35.17%and 75.80%compared with the original signal.2.In the feature extraction stage,the commonly used feature points with clear physiological significance in time domain and the feature parameters related to each feature point are extracted.On this basis,the height of the main peak is divided into ten equal parts,and the sample point on the corresponding falling edge is found,and the slope,distance and area from the sample point to the next starting point of the pulse wave are calculated as new features.All the features are analyzed,the results show that the Mean Absolute Error of the model prediction is smaller when the new feature subset is added than when the common feature parameters are used alone.Using BP neural network to establish systolic blood pressure and diastolic blood pressure network models for PPG signals after artifact removal,the mean absolute error of systolic blood pressure is 4.26 mmHg and the standard deviation is 7.49 mmHg.The mean absolute error of diastolic blood pressure is 4.15 mmHg and the standard deviation is 7.01 mmHg.The prediction results of blood pressure after adding new features to the PPG signal after artifact removal all meet the detection standard of A AMI.