Compressed Sensing And Intelligent Reconstruction Of Heart Sound Signal

Heart sound is the sum of various mechanical vibrations produced when blood flows in the cardiovascular system,and has characteristic elements such as amplitude and period.Usually the vibration frequency of the heart sound signal is between 20Hz and 800Hz,which is within the range that our human ears can hear.Due to the mechanism of heart sound generation,heart sound signals carry a lot of information about cardiovascular health.In clinical practice,heart sounds are the main source of information used to assess heart function.With the birth of the Internet of Things and 5G technology,mobile medical and smart medical services have also been rapidly developed.People's requirements for data are gradually increasing,and the amount of data that needs to be processed is also gradually increasing.Not only is the data required to be accurate,but also faster and more efficient storage.Compressing data,reducing the amount of data transmitted,and reducing communication consumption are one of the important directions for large-scale data sharing in the future.When the heart sound signal needs real-time transmission,it is necessary to reduce the delay as much as possible and improve the compression efficiency.Because the heart sound signal is auxiliary diagnosis,its accuracy is also very high,that is to say,it requires high distortion after compression and reconstruction.Other medical signal or image compression similar to heart sound also has the same problem.This thesis only uses heart sound as the beginning of a trial study of a new compressive sensing and intelligent reconstruction method.In this context,this thesis mainly applies two new methods to realize the practice of signal compression: 1)A new method of signal decomposition and reconstruction — Discrete Convolutional Wavelet Transform(DCWT).2)Improved clustering method based on K-means clustering—clustering method based on correlation.Discrete convolutional multiwavelet transform can decompose and reconstruct the signal in real time.The principle is simple and does not require a particularly complex mathematical theory basis.You only need to find the necessary conditions for reconstruction and the appropriate selection of the number of filters and the length of support.A simple technical framework for signal processing decomposition and reconstruction.The correlation-based clustering algorithm solves the traditional K-means clustering algorithm,which is sensitive to abnormal points and difficult to select the number of clustering categories k value.It can perform discrete convolutional multiwavelet decomposition of the signal well.After clustering,the clustered signals are finally quantified,and each class is assigned a small or even super low bit rate to achieve the purpose of compressing the heart sound signal and rebuilding it back.This is a new compressed sensing method.Through selective clustering of the heart sound feature values,each class only needs to store the class label or some quantitative features,so as to achieve the purpose of compressing and reconstructing the heart sound signal.In order to achieve better results after reconstruction,the idea of total variation filtering is used to intelligently restore and reconstruct the heart sound signal after lossy compression.The main innovation of this paper is to construct a new signal compression and reconstruction method by using DCWT signal decomposition and reconstruction method and clustering method.Therefore,the main technical framework of DCWT,correlation clustering method,quantization coding,total variation filtering and other technical implementation details are also introduced.