A Research On Heart Sound Signal Processing For Wearable Applications

Wearable heart sound monitoring is an emerging research topic for cardiovascular diseases.However,most of the current heart sound monitoring devices only have data storage and transmission functions,and the wearable applications are susceptible to interference,and there is a lack of matching algorithms to realize the processing of heart sound signals for wearable applications.In order to solve the above problems,this paper designs and implements noise reduction,segmentation and classification algorithms for wearable heart sound monitoring applications,and the main research contents are as follows.1)Design a noise reduction algorithm for wearable heart sound monitoring applications.The algorithm combines Variational mode decomposition(VMD),GHM multiwavelet and Principal Component Analysis(PCA)to achieve noise reduction of the original noisy signal according to the characteristics of the heart sound signal and the noise features of the wearable application.(This method is referred to as VGP in the following).In the noise immunity test on the public data set and hardware simulation platform,the heartbeat signal can be processed stably in a noisy environment with a signal-to-noise ratio of-5d B,and the signal-to-noise ratio can be improved by more than 5d B at 5d B.2)Design an S1 and S2 segmentation and classification algorithm for wearable heartbeat monitoring applications.After completing the noise reduction of the heart sound signal,the segmentation of S1 and S2 is performed using a double threshold method based on the Teager energy operator and the higher-order Shannon envelope,and finally the types of S1 and S2 are judged using SVM.Validation on a publicly available dataset yields an average check-all rate of 95.5% for S1 and S2 detection and97.2% for classification.The localization error is less than or equal to 6.3% in a noisy environment with a signal-to-noise ratio of-5 d B.3)Design a processing algorithm for S3,S4 and heart murmurs for wearable heart sound monitoring applications.Since the incidence and amplitude of S3,S4 and heart murmurs are low.Therefore,the presence detection of S3,S4 and heart murmurs needs to be carried out separately from the segmentation steps of S1 and S2,and a set of parameters and processes for localization and classification are adopted independently according to their time-frequency domain characteristics.In summary,the heart sound signal processing algorithm for wearable heart sound monitoring application proposed in this paper is capable of stable segmentation and classification of each component of heart sounds in complex noise environment.