Research On ECG Feature Extraction And Classification Algorithm Based On Kernel Independent Component Analysis

With the gradual improvement of people’s living standards, people’s awareness of health is also increasing. However, the incidence of heart disease which has been gradually to "the first killer" as a threatment to human health is on the rise year by year, so it is extremely important to master the occurrence and development rule and its prevention and control measures. Electrocardiogram (ECG) is a reflection of the cardiac electrical activity in the body, and it has an important guiding significance for the detection and diagnosis of heart diseases in clinical. Because of the complexity of ECG signal waveform and the interference of various noise, it is difficult to extract effective data features from the ECG signal. Therefore, it has important theoretical and practical value to study the feature extraction and classification algorithm of ECG signal.According to the characteristics of the ECG signal, the thesis presents a improved algorithm based on kernel independent component analysis in combination with discrete wavelet transform to extract the features of the ECG signal. First, the 20 dimensions features are obtained by using principal component analysis to reduce the dimension of the sample data which is used to extract the nonlinear feature of the kernel independent component analysis. Second, through the discrete wavelet transform to extract detail coefficients of the first scale to the fourth scale and the approximate coefficients of the fourth scale as a frequency domain features, using statistical methods obtain 20 dimensions after taking its maximum, minimum, mean and standard deviation respectively, then using linear discriminant analysis optimization for 4 dimensions. Finally, the features of optimization are composed of the multiple domain feature space. In the classification diagnosis of support vector machine classifier design, the thesis selects LIBSVM as the classifier for the optimized feature vector classification, and the genetic algorithm is adopted to optimize two parameters of the LIBSVM:the penalty factor C and radial basis function (RBF) kernel width g. Then five types of signals of MIT-BIH arrhythmia database are classified:normal signal, left bundle branch block, right bundle branch block, premature ventricular contraction and artial premature beat. The performance of the classifier has the sensitivity, specificity and positive predictivity accuracy, and their average value are 98.50%,99.69% and 98.91% respectively. In addition, the classification accuracy of the testing set is 98.8%, which achieves the desired classification results. In the end of this thesis, the proposed algorithm is used to classify the ECG signals collected from the Prosim 2 vital signs simulator.In this thesis, by extracting the essential features of ECG signal accurately, we can realize the classification and recognition of different types of ECG signal with high accuracy. It has a positive significance to improve the diagnostic efficiency, shorten the time of diagnosis, condition monitoring for the patients with heart disease and have an effective rehabilitation evaluation.