Research On ECG Feature Extraction And Classification Method

The subject of automatic diagnosis of ECG signals has great theoretical signif-icance and practical value. This thesis mainly attempts to develop a methods of au-tomatic ECG diagnosis with high precision. In order to develop a reliable automaticECG diagnosis system, we need to deal with a number of critical issues covering fromsignal acquisition to diagnosis results.To achieve such a practical system, we start from the study of preprocessing ap-proach of the ECG signals. It involves many common issue of dealing with clinicalsampling, including removal of different kinds of noise and baseline drift. We test andevaluate available signal preprocessing on many approaches and select the best oneimplemented in our online diagnosis system. We propose a statistical beat segmenta-tion approach and a wave peak detection method base on multi-lead information, sothat we can estimate the beat section more accurately.It is an important question that how to map the biology signals to the feature s-pace. We combine the time domain feature and statistical feature as the final featureto classify. Principal Component Analysis and Independent component analysis ap-proaches are respectively used on feature extraction experiments. It shows that theICA basis have higher ability as feature extraction basis in classification.To achieve high accurate prediction results of multi-category classification, wedesigned a vote strategy of multiple classifier, which combined of probability estima-tion. Finally we get the higher test classification accuracy result than former research-es.