ECG Classification Based On Sparse Representation And LS-SVM

Cardiovascular disease is one of the main diseases which harm human health. The latest statistics promulgated by World Health Organization showed that chronic diseases caused increasing numbers of deaths worldwide. Moreover, the ischemic heart disease which caused700 million(12.9%)deaths occupied the first place among the 10 leading causes of death in2012. Most clinicians analysis and diagnosis of disease through the ECG waveforms, and automatic analysis of ECG signals is undoubtedly one of the key technologies for the ECG monitor today. The task of ECG automatic analysis mainly includes the ECG signal preprocessing, feature extraction, classification and so on.However, owing to the complexity and the diversity of ECG signals, it is very difficult to classify accurately. Existing ECG classification algorithm has made many achievements,but further studies are required both in theory and application. The emergence of sparse representation provides new point of penetration for ECG signal processing. Sparse representation has been successfully applied to face recognition, image denoising, ECG compression, ECG detection, etc. A new method that combines sparse representation with Least Squares Support Vector Machine-based classification is proposed for the classification of five types of ECG. Research of the dissertation is concentrated on the following aspects:(1)ECG sub-dictionaries were constructed by FCM, so that the atoms as similar as possible intra-class and as different as possible between inter-class;(2)Sparse coefficients were solved by each sub-dictionary instead of by a large dictionary consists of all sub-dictionaries, such manner not only improves the classification accuracy but also increase efficiency;(3)Multi-feature fused sparse coefficients by LS-SVM, both ensure exploit the characteristics of the data fully and improve the generalization ability.