Research On ECG Feature Fusion And Feature Optimization Algorithm Based On Improved EEMD And GA

The high rates of death and disability caused by arrhythmias place a heavy burden on families and society.The effective prevention and diagnosis of arrhythmias has become an urgent issue.Among the many methods available,real-time automated monitoring of the human electrocardiogram(ECG)is one of the most effective means of preventing and diagnosing arrhythmias.The accuracy of arrhythmia identification is one of the most important issues in automatic ECG monitoring,and ECG signal feature extraction is a key step in improving the accuracy of the identification.However,the existing ECG signal feature extraction algorithms are not deep enough and comprehensive enough,and the algorithms are not very generalized.To address these problems mentioned above,this thesis proposes an ECG signal feature fusion and feature optimization algorithm based on improved ensemble empirical mode decomposition(EEMD)and genetic algorithm(GA)with the research objective of improving the ECG signal classification performance,and the main research contents are as follows.(1)ECG signal pre-processing.The ECG signal in the MIT-BIH heart rate abnormality database was first denoised using a finite impulse response filter,then the dual-slope feature extraction algorithm was used to extract the dual-slope features for the R-wave peak localization of the ECG signal,then the heartbeat segmentation was performed based on this localization result,and finally the segmented heartbeat signal was amplified using the sliding window integration method.The ECG signal is then amplified using the sliding window integration method.The ECG signal is then used to provide data support for subsequent feature fusion and feature optimization.(2)A time-frequency domain feature fusion algorithm for ECG signals based on improved EEMD and statistical analysis.The algorithm first improves the EEMD algorithm using principal component analysis on the frequency domain feature extraction of ECG signals.Then,on the time-domain feature extraction of ECG signals,the features in terms of amplitude features,low-order statistics,high-order statistics and entropy values of ECG signals are extracted based on the principle of statistical analysis.Finally the extracted time-frequency domain features were fused.The Support Vector Machine(SVM)classification validation experiments showed that the overall classification accuracies were98.54% and 93.84% for the intra-patient and inter-patient cases respectively.(3)An improved GA-based feature optimization algorithm for ECG signals.The algorithm uses the normalized Euclidean distance as the fitness function suitable for the feature optimization algorithm in this thesis for the calculation of the global optimal solution by improving the fitness function of GA,which makes the feature extraction algorithm more universal.the SVM classification validation experiments show that the overall classification accuracy is 98.77% and 94.19% in the intra-patient and inter-patient cases,respectively.This thesis improves the ECG signal feature extraction algorithm from the perspectives of feature fusion and feature optimization,and improves the generalizability of the algorithm while ensuring high classification accuracy.The proposed algorithm is conducive to the application of feature extraction algorithms to the automatic diagnosis and identification of cardiac arrhythmias,and provides a certain reference for promoting academic research on ECG signal feature extraction as well as promoting its application.