The Research on Feature Extraction Method of ECG Signal Based on KPCA Dimension Reduction

Electrocardiogram(ECG) contains abundant human body information and plays an important role in heart disease analysis. At present, the analysis of ECG is a quantitative or qualitative analysis of the amplitude or time interval of the relevant feature points mainly based on its time-domain waveform, and the extracted feature dimensions are high and contain redundant information. This paper first extracts 7 linear features and 9 nonlinear features from the ECG signals, then uses the KPCA algorithm to extract low-dimensional principal component features from the high-dimensional original feature space. The experimental results show that the extracted original features have significant statistical difference between normal rhythm group and arrhythmia group (p<0.05), which provides a basis for subsequent principal component features extraction, and when only 5 principal components are retained, its cumulative contribution rate exceeds 90%, and the effect is better than the PCA algorithm.