Research On Feature Extraction And Recognition Of CHD Heart Sound Based On CEEMDAN Arrangement Entropy

Cardiac signal is an important physiological signal of the human body.Cardiac auscultation is the main method for the initial diagnosis of congenital heart disease.However,auscultation mainly depends on the experience of auscultation experts.However,relying on doctors to auscultate will result in inaccurate judgment results,and misdiagnosis will inevitably occur.Cardiac signal preprocessing,feature extraction,classification and recognition are conducive to the diagnosis of congenital heart disease.This paper hopes to reveal the potential features of congenital heart disease through signal processing,traditional pattern recognition and machine learning.The foundation for clinical diagnosis.The research work of this paper is mainly the following three parts:1.Preprocessing of heart sound signals.Aiming at the problem that the collected heart sound signal will have noise and how to filter the cardiac cycle signal,a denoising method based on the new threshold function of lifting wavelet is proposed.In order to better screen the effective cardiac cycle signal in the heart sound signal,a A method of combining the normalized Huanian energy Viola integral waveform is used to extract the envelope of the denoised heart sound signal,and then use the double threshold method for segmentation positioning processing to determine the cardiac cycle of the heart sound,and extract the signal for each case.After the cardiac cycle is screened for a weighted average,the final cardiac cycle signal is confirmed.2.Feature extraction of heart sound signals.Aiming at the non-stationary and nonlinear problems of heart sound signals,a feature method based on Complement Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)and extraction entropy is proposed.The signal is decomposed into several Intrinsic Mode Function(IMF)components from high frequency to low frequency.The C C algornthm is used to optimize the parameters in the permutation entropy of each IMF component.Finally,the IMF components of each order are obtained.The entropy is arranged and constructed into a feature vector.3.Classification and recognition research:According to the sparsity characteristics of the extracted permutation feature vector data,a classification method based on depth factorization machine network is proposed to classify normal and abnormal signals,using loss rate(LOSS)and accuracy.(Accuracy)and AUC(Area Under ROC Curve)three indicators to measure the generalization performance of the depth factoring machine network model.In this paper,600 normal heart sound signals and 600 congenital heart sound signals are studied and analyzed.The experimental results show that CEEMDAN and permutation entropy are used to extract the heart sound features and feature vectors,which can better reflect the different frequencies contained in the heart sound signals.Characteristic component.The heart sound signal is classified and identified by the depth factorization machine network,and the loss rate is 0.213,the accuracy rate is 0.887,and the AUC value is 0.902.