Research On Filtering And Feature Recognition Method For Electrocardiogram Signal Under Strong Noise Background

Cardiovascular disease is a serious threat to human health,which is one of the most concerned diseases in the society.As a signal that can effectively reflect heart activity,Electrocardiogram(ECG)is an important reference in the diagnosis of cardiovascular disease.In recent years,how to remove the strong noise interference of ECG,detect the characteristic waveform and recognize the arrhythmia are the research hot spots.In order to provide help for ECG processing and analysis,this thesis aims to develop a filtering algorithm under strong noise background,propose an improving QRS wave detection algorithm and establish a more effective arrhythmia classification model.The specific research work is as follows:Firstly,the research of ECG filtering algorithm under strong noise background is carried out.According to the characteristics of various noises,a suitable noisy signal model is established.Next,an adaptive ensemble empirical mode decomposition(EEMD)algorithm is proposed and the determination criteria of key parameters are optimized.Referring to the idea of wavelet threshold denoising,the improved threshold function and new threshold are proposed to eliminate high-frequency noise.The soft threshold function and fixed threshold are used to eliminate low-frequency noise.It is proved that the algorithm can effectively promote SNR,reduce RMSE and maintain waveform features.The comparative analysis shows that the algorithm in this thesis has stronger ability to remove strong noise.Then,the research of QRS complex detection algorithm is studied.According to the characteristics of waveform,the research is divided into two directions: R wave detection and Q-S wave detection.R wave detection algorithm based on variable mode decomposition(VMD)is proposed.The determination criterion of preset scale is improved.Combining with the extraction of envelope peak value,the work of R wave location is completed.It's verified that the overall recognition rate of the algorithm is99.71%.According to the characteristics of Q wave and S wave,on the basis of R wave accurate positioning,the algorithm of window search minimum is used to detect Q wave and S wave.The principle of this method is simple and its computation load is small.Finally,the arrhythmia classification of N,V,A,L and R is carried out.According to the characteristics of ECG samples in MIT-BIH database,a one-dimensionalconvolutional neural network(CNN)classification model is designed.27500 beats data are selected for the experiment.The training and testing are completed on Matlab with the method of ten-fold cross validation.The results indicate that the overall recognition rate of the model is 98.5%,which is superior to the four algorithms of the control group.Then the analysis of each index shows that the algorithm has poorer effect on the recognition of class A beats.In the end,the setting principle of two important parameters in the experiment is discussed,which provides support for the scientific design of the model.