Research On The Arrhythmia Classification Methods For The Wearable ECG Monitoring Modules

With the development of science and technology and people's attention to the health field,wearable physiological monitoring devices are developing rapidly.Wearable monitoring devices have achieved a good result for some common,prone,chronic diseases,which are convenient for people.In the context of remote medical and Internet,it is easier to achieve family medical and disease prevention.Cardiac diseases are a threat to people's health in some chronic diseases,therefore,it is very meaningful to propose an arrhythmia identification and classification method which is more suitable for wearable ECG monitoring module.This thesis uses the mature method to achieve ECG signal denoising and ECG wave detection.The thesis extracts the multi-domain features,morphological features,wavelet domain features,cyclic spectral features and high-order cumulant features.The morphological features and wavelet domain features have been widely used in the classification of arrhythmia and achieved good results.Two common classifiers,support vector machine and extreme learning machine,is selected to validate the effectiveness of cyclic spectral features in arrhythmia classification tasks.Experimental results further prove that the cyclic spectral feature can extract the hidden information and is more suitable for arrhythmia classification tasks.Most of the conventional methods for arrhythmia classification either observe single feature of ECG signals or just make a linear fusion for all features,which is not enough to distinguish multiple types of arrhythmias.Therefore,capturing local information is more preferable.Usually,the classification accuracy depends on the distribution of samples.In this thesis,a subspace-based method for the classification of arrhythmias is proposed.Besides,a cascaded classifier based on a Support Vector Machine is constructed to classify arrhythmias.Experimental results show that the optimal feature subspace method has achieved a better classification performance and completed the multi-classification task.