Research And Design Of Heart Sound Signal Classification And Recognition System Based On FPGA

Research and analysis of heart sound signals for non-invasive diagnosis of cardiovascular disease has irreplaceable important role.With the development of digital signal processing technology,research and application of heart sound from physician's subjective auscultation of heart sound signal to digital heart sound classification system,the system assisted the doctor more efficient diagnosis of cardiovascular disease and improve the accuracy of diagnosis.But with the development of medical device miniaturization of portable,the new require of fast computing and easy to carry is presented.The paper design a heart sound recognition system based on FPGA,hardware analysis of heart sound signals is very faster,and more convenient to carry.This design uses the hardware FPGA chip is Xilinx XC5V1X30 chip,software environment for the ISE software,acquisition module in homemade circuit boards.The paper main work is as follows:1)The system uses strong anti-jamming ability of heart sound sensor converts the vibration of signal of the heart to the voltage;Use INA337 precision instrument amplifier for the amplify voltage signal,and then through the bandpass filter circuits,to AD(AD9268)converter analog input voltage requirements,AD convert into FPGA chips processing system.2)In the FPGA chip design a sound signal denoising module based on verilog language.Select Db6 Wavelet for decomposition of heart sound signal 5 layer decomposition,and selecting the appropriate threshold values for each layer of the high frequency coefficient to threshold process,then reconstructed to denoising of heart sounds.3)Extracted the characteristic parameter of denoising heart sound signal.At first,EMD decompose the denoising heart sound signal to get five IMF,then Hilbert transform on this 5 IMF component can get instantaneous frequency and amplitude envelope.Then using Wavelet packet decompose the denoising heart sound signal and computing energy spectrum of every Wavelet packet decomposition coefficient.All Modules include EMD decomposition module,Hilbert transform module,wavelet packets decomposition module.EMD decomposition by soft core implemented in the FPGA chip,using c-language implemented in the soft core EMD decomposition method;and modeling in the System Generate a Hilbert transform and Wavelet packet analysis module,and then converted to a hardware description language.4)Finally,classification and recognition modules,the paper select BP neural network as the classification and identification system.Neural network module for submodules,including incentive function module,the hidden layer neurons module and output neurons module and the control module.Modeling of neural network module to the System as a whole Generate design,each layer of the weights obtained by Matlab off-line learning.The paper selected mitral stenosis heart sound signal,mitral valve insufficiency of heart sound signal,Atrial Septal Defect heart sound and normal heart sound four types of heart sound recognition.Experimental results show that the system can identify these four types of heart sound signal,recognition rate is above 96%.