基于NAR-KF的心音信号仿真研究

Research on heart sound signal denoising based on NAR and Kalman filtering

为了建立一维时间序列的心音信号模型,通过医院采集数据,采用非线性自回归(NAR)神经网络对S1与S2心音信号进行建模,在得到心音信号的预测值后,对心音信号使用卡尔曼滤波方法进行降噪。为验证融合算法对于心音信号降噪的可行性与优越性,进行了一系列仿真实验。在同时考虑精度与训练时间的情况下得到了一组较为理想的模型,再将该模型输入卡尔曼滤波的预测值中,通过原心音信号进行滤波,对比降噪前的第一心音信号滤波值的均方误差,有较为优越的降噪性能。得到的第二心音信号对比降噪前也有较为明显的提升。实验结果表明,融合算法在信噪比以及均方误差等降噪性能上有明显的优越性。

In order to establish the heart sound signal model with one-dimensional time series,the nonlinear autoregressive(NAR)neural network is adopted to establish the models of the heart sound signals S1 and S2 by the data collected from a certain hospital.After the predicted values of the heart sound signals are obtained,the heart sound signals are denoised by the Kalman filtering method.A series of simulation experiments were carried out to verify the feasibility and superiority of the fusion algorithm for denoising of the heart sound signals.In the case of considering the accuracy and training time at the same time,a group of ideal models were obtained,and then the model values were input into the predicted values of the Kalman filtering to execute filtering by means of the original heart sound signals.In comparison with the mean square error(MSE)of the filtering value of S1 before noise reduction,the fusion algorithm has a better noise reduction performance.The quality of S2 obtained is also significantly improved in comparison with that before noise reduction.The experimental results show that the fusion algorithm has obvious advantages in noise reduction property including signal-to-noise ratio(SNR)and MSE.