Research On Denoising And Classification Algorithm Of Heart Sound Signal

In recent years,cardiovascular disease(CVD)has seriously endangered people’s physical and mental health.Heart sounds carry all kinds of information that the human body maintains the normal operation of the heart,so the analysis of heart sound signals is of great significance for the prevention and diagnosis of CVD.The traditional method of heart sound diagnosis is manual auscultation,but there are certain subjectivity and limitations,so the accuracy is not high.If a set of algorithms can be developed to analyze the heart sound signals and then automatically classify,it will be able to assist doctors to make judgments on diseases.Combined with this need,this paper mainly studies the following aspects.De-noise the heart sound signal.In order to effectively remove the noise mixed in the heart sound signal,an improved wavelet threshold denoising algorithm is designed.Select db6 as the wavelet basis function,the number of decomposition layers is 5 layers,retain useful decomposition layer coefficients,and set the coefficients of the remaining layers to zero.In addition,a new adaptive threshold estimation method and threshold function are designed.Parameters are introduced to reflect the noise level in the signal.According to its adaptive estimation threshold,the new threshold function is used to reconstruct the signal after thresholding the coefficients.Under different noise intensities of different types of noise,the soft threshold,hard threshold and new threshold functions are used to denoise the heart sound signal.The experimental results show that the designed algorithm has the best denoising effect.Perform feature extraction on heart sound signals.Short time energy,short time average zero crossing rate and short time autocorrelation function are selected as three time domain features.The Mel Frequency Cepstral Coefficient(MFCC)method is used to extract 12-dimensional MFCC parameters as 12 frequency domain features.The instantaneous frequency of heart sound is obtained by Hilbert-Huang transform(HHT),and its mean and standard deviation are extracted as two frequency domain features.The 17 time-frequency features are mixed as the feature vector of the heart sound signal and used as the input of the later heart sound classifier.The multi-feature parameter mixing method can improve the classification accuracy.Classify and recognize heart sound signals.Using BP neural network as the classifier,an improved particle swarm optimization and improved BP neural network(PSO-BP)algorithm is designed.Aiming at the problem that the traditional BP neural network is easy to fall into the localminimum value,it is improved by adding the momentum term and adaptively adjusting the learning rate.In view of the problems of premature convergence and slow convergence speed of traditional PSO algorithms,the three aspects of PSO’s inertial weight,learning factor and convergence accuracy are improved.Finally,the improved PSO algorithm is used to optimize the initial weights and thresholds of the improved BP neural network,and an improved PSO-BP neural network is obtained.The traditional BP network,the improved BP network and the improved PSO-BP network are used to classify the heart sound signals.The experimental results show that the designed algorithm has the fastest convergence speed and the shortest training time,and the best classification effect.