ECG Classification Based On Sparse Coding

With technology and science developing,the standard of living of human beings has increased year by year.This brings mental problems such as unhealthy diet and high pressure from work resulting in an increase in the incidence of cardiovascular disease(CVD)year after year.A survey shows that cardiovascular disease has become one of the major causes of human abnormal death in the world.Electrocardiogram(ECG)is an effective indicator used by modern medicine to check the physical condition.ECG monitor has become an effective medical device to monitor patients' heart disease in many hospitals and private medical services.ECG signal analysis is an important technology in the ECG monitor.A good ECG signal analysis system needs an effective algorithm to perform signal processing on the collected ECG signals for the follow-up arrhythmia detection.ECG automatic analysis and detection technology includes the pre-processing technology of ECG signals and ECG signal feature extraction technology,etc.To obtain the highest possible accuracy in the final detection of arrhythmia is our ultimate goal.At the same time,it is expected to take into account the pressure on hardware requirements when the module is implemented,signal storage and processing speed and more are all we want to solve.Because of the diversity and complexity of human electrocardiogram signals,it is difficult to accurately classify ECG signals.Since ECG signals can be sparsely expressed,this can serve as a new entry point for scientific research.Sparse representation has been widely used nowadays,especially using sparse representation to the high-dimensional data,and it has become an important topic in the field of machine learning and AR vision research.In this paper,we ues sparse coding to process the ECG signals.We uses K-SVD algorithm in sparse coding.Through pre-processing and heartbeat segmentation of the original ECG signals,a single heartbeat dataset is obtained.Using wavelet transform and independent vector analysis,the wavelet feature and ICA feature matrix of the heartbeat signal are extracted,and the Overcomplete dictionary of the heartbeat signals are established by K-SVD algorithm using sparse coding.experiments show that the algorithm has good separability for the characteristic wave of ECG signal,and the sparse coding is of high fidelity.The sparse representation of ECG signals was used in the early stage to use a clustering algorithm to make the atoms between dictionaries as different as possible,and the atoms in the dictionary are as similar as possible;the main purpose of sparse representation is to seek the optimal approximation of the signal in a certain space.In the transform domain,the original signal is expressed with as few atoms as possible,so that the basic information of the original signal can be grasped as a whole.This method also improves the overall recognition rate of a variety of arrhythmia signals,and has better performance for the storage and classification of large-scale heart beat data.This paper uses sparse coding and Support Vector Machine(SVM)to classify ECG signals.Support vector machine(SVM)is a two-class classifier that performs feature extraction on the original ECG signal and reduces the training dimension of the signal.This paper selects the improved model of support vector machine hierarchical support vector machine.The vector machine is a kind of binary tree SVM.A part of the data set is selected as the training set.The training model of the classifier is obtained through training the training set using the SVM algorithm.Finally,the heartbeat classification was performed on the six categories of ECG signals,and the accuracy rate reached 95.16%.Experiments show that the hierarchical support vector machine classifier can have good classification performance on the processed ECG signals.