Feature Extraction And Classification Of ECG Signal Based On EEMD And Fractal Theory

With the continuous development and progress of society,residents’ lives have entered a fast pace,the pressure in all aspects has increased,and more and more people have developed irregular eating and living habits,so that the body is in a high-load operation state for a long time.And with the aging of the population,the incidence,disability rate and mortality rate of cardiovascular diseases continue to be high,which poses a huge threat to the life and health safety of residents.Arrhythmia is one of the important manifestations of cardiovascular disease,but arrhythmia is often not easy to detect,once there is obvious discomfort,usually the disease has reached a serious stage,and there is a possibility of sudden death and heart failure at any time.As a simple and non-invasive medical test,electrocardiogram occupies an important position in the detection of heart disease.However,in the traditional medical diagnosis and treatment methods,most clinically experienced doctors are used to determine whether the test person has heart disease through ECG,and longterm identification work will make doctors have visual fatigue,and the probability of missed and misdetected tests will increase.With the further development of computer technology,the cost of manual recognition can be reduced with the help of the powerful computing power of computers,and the accuracy of ECG signal classification can be improved.On the one hand,the current classification method tends to classify the frequency domain,wave domain,and morphological characteristics manually,and on the one hand,it is classified by relying on the characteristics of automatic learning of ECG signals in neural networks.These methods are difficult to study ECG signals from a nonlinear perspective.The thesis is based on normal beat,atrial premature beat,ventricular premature beat,left bundle branch block beat and right bundle branch block beat as research objects,conduct nonlinear analysis,and study the classification method of ECG signals.The main research contents and achievements of the thesis include the following points.(1)Denoising of ECG signals.Ensemble empirical mode decomposition was used to decompose the ECG signal into 6 intrinsic mode functions and 1 residual term,calculate the correlation coefficient between each intrinsic mode function and the original signal,and select the appropriate intrinsic mode function to reconstruct the ECG signal.(2)Multifractal feature extraction and classification of ECG signals.From the perspective of nonlinear of ECG signals,the multifractal characteristics of ECG signals are studied,and experiments show that the multifractal characteristics of ECG signals can be enhanced by ensemble empirical mode decomposition.The mass index curve,generalized fractal dimension and multifractal spectrum of ECG signals are studied,and suitable multifractal features are extracted for the training of support vector machines.The average classification accuracy obtained by 30 tests with multifractal feature training was 96.09%,and the classification accuracy of normal pulsation and left bundle branch block signals in a single experiment could reach more than 97%,which proved the effectiveness of this method in ECG signal classification.(3)ECG signal classification method based on fractal fusion feature.On the basis of multifractal analysis,the multifractal correlation features of ECG signals were further studied,and three multifractal correlation features of singular index maximum,multifractal correlation spectral width and asymmetric index were extracted and fused with five multifractal features,and then combined with SVM classification.The experimental results show that when the fractal fusion feature is used for experiments,the highest classification accuracy in the five-fold cross-validation can reach 98.67%,and the accuracy of each fold reaches 97.8%,and the average accuracy of the training set and the test set can reach 98%.