Research On SVM Heart Sound Signal Classification Algorithm Based On Optimization

Although the accuracy of the medical heart sound detection device has been continuously improved,heart sound signal still have noise mixed due to the influence of surrounding environment and other physiological activities of the human body,which affects the medical diagnosis of the heart disease of the patient.According to the problems above,this paper researches the heart sound signal classification algorithm based on optimized support vector machine(SVM).In this paper,the empirical modal decomposition algorithm is used to analyze the feature distribution between heart sound signal and noise.It is concluded that the heart sound signal is mainly concentrated in the low frequency part,and the noise is mainly concentrated in the high frequency part.According to the feature distribution,the combination of the Chebyshev filter type II low-pass filter and the spectral subtraction algorithm is used to denoise the collected heart sound signal data.Among them,the Chebyshev filter type II low-pass filter mainly reduces the noise in the high-frequency heart sound signal part.After the heart sound signal passes through the low-pass filter,the high-frequency signal energy will decay.Spectral subtraction algorithm is used again for the heart sound signal after noise reduction,in other words,the spectral energy of the total heart sound signal subtracts the spectral energy of the noise signal,and the inverse Fourier transform is used,so that the time series of the heart sound signal after the noise reduction can be obtained.Next,combined with the electrocardiogram of the heart sound signal,it can be known that the waveform of the normal heart sound signal is composed of three waves of P wave,QRS wave and T wave.When the heart is abnormal,its waveform will change in time and period.According to the phenomenon above,this paper uses the Mel cepstrum coefficient algorithm and the cepstrum pitch detection method to extract the features of the heartbeat signal after noise reduction.Because of the problems ant colony clustering algorithm brought that make the SVM's penalty factor and kernel function parameters fall into the local minimum,the convergence speed is slow and so on,in this paper,considering the shortcomings of traditional ant colony clustering algorithm,the traditional ant colony clustering algorithm is optimized by adding particle swarm algorithm,which solves the problem that traditional ant colony clustering algorithm has slow convergence speed and falls into local optimum.The improved SVM firstsets its predictive classification function as the objective function.The particle swarm algorithm uses the objective function to find the individual local extremes and global extremum of the particle according to the initial value and range of the penalty factor and the kernel function parameters.By setting the initial cluster center,the ant colony clustering algorithm is based on the initial cluster concentration and pheromone emitted by each ant,it could training individual population extremes and global extremum samples of particle swarms algorithm,and find the optimal value of the local and global extremum parameters.Experiments show that the particle ant colony clustering support vector machine algorithm is easier to find the optimal value of the prediction classification function in the sample than the traditional SVM,and the heart sound signal classification is more accurate.The improved SVM algorithm has an accuracy of 95.4023%,which is an increase of 9.1954% compared with the traditional SVM.At the same time,it can shorten the training time and speed up the convergence.