ECG signal classifying method based on wavelet packet and approximate entropy

Abstract

The invention discloses an ECG signal classifying method based on a wavelet packet and approximate entropy. The method comprises the steps that wavelet packet decomposition is conducted on preprocessed ECG signals, the wavelet packet coefficients of all nodes are extracted, and then the approximate entropy of all the wavelet packet coefficients is calculated; the obtained approximate entropy is used as a feature vector to be input into a support vector machine classifier, a particle swarm algorithm is used for seeking the optimal parameter of the support vector machine classifier, and various ECG signals are classified. Wavelet packet decomposition is an effective method for analyzing non-stationary signals. The approximate entropy can be used for obtaining stable estimation values only by a small amount of data, good noise-proof and anti-interference capacity is achieved, and the approximate entropy can both be used no mater what the signals are, stochastic or deterministic. According to the ECG signal classifying method, an algorithm for extracting feature vectors is simple, dimensionality reduction in a traditional method is of no need, the speed is high, consumed time is small, the classifying accuracy is high, and the ECG signal classifying method is suitable for an ECG automatic analysis auxiliary diagnosis system.

Description

technical field [0001] The invention belongs to the technical field of medical signal processing, and relates to a method for classifying electrocardiographic signals based on wavelet packets and approximate entropy. Background technique [0002] The electrocardiogram is a comprehensive reflection of the electrical activity of the heart on the body surface, and provides an important reference for diagnosing the functional changes of the heart and cardiovascular system. ECG pattern classification mainly depends on the correct extraction of ECG signal feature vectors and the accuracy of classification methods, which is an important part of automatic analysis and diagnosis of ECG signals. At present, many scholars have conducted research on this, mainly involving two aspects of feature extraction and classification methods of ECG signals. Fourier transform is simple and accurate in extracting time-frequency domain features, but it cannot accurately reflect the morphological ch...

AI Technical Summary

Problems solved by technology

Fourier transform is simple and accurate in extracting time-frequency domain features, but it cannot accurately reflect the morphological characteristics of ECG signals, and is easily disturbed by noise; although feature extraction such as KPL and wavelet transform have their own characteristics, their accuracy, anti-interference, etc. It has a shortpart

The traditional feature extraction mainly uses the linear transformation method to extract a single feature of the ECG signal, which cannot completely reflect the nonlinear structure of the ECG signal. Although principal component analysis and independent component analysis can extract the nonlinear features of the signal, they need to extract a large number of features, and then perform dimensionality reduction selection, which is time-consuming

The neural network method is difficult to establish the optimal network structure, the environmental adaptability is not high, and there may be problems of over-learning or easy to fall into local minimum points, and the generalization is not strong

Images