心电信号平滑分解阈值去噪方法

The denoising effect of wavelet decomposition depends on the choice of wavelet functions. Empirical mode decomposition has the problem of mode aliasing, and the ensemble empirical mode decomposition requires a lot of repeated decomposition process. In order to solve the above problems, a smoothing decomposition threshold denoising method based on the principle of smoothing filters is proposed for denoising ECG signals. Firstly, the noisy ECG signal is decomposed into several smoothing decomposition components from high frequency to low frequency, and the boundary threshold is established to divide the components into three groups: the high-frequency noise components group, the useful signal components group and the low-frequency noise components group. The time-varying threshold and the interval threshold are proposed to eliminate the high-frequency noise in the first group. The ECG signal is reconstructed by adding the denoised high-frequency noise components and the useful signal components. The denoising experiment results for the simulated signals and the real signals from MIT-BIH database show that the signal-to-noise ratio (SNR) within the experimental range is improved by 19.656 dB~2.448 dB. The correlation coefficient is between 0.885 and 0.999 and the energy ratio is close to 1. P waves, T waves and QRS complexes of the denoised ECGs are complete and clear. The position, shape and amplitude of the waveforms are consistent with the source signals, indicating that the proposed method has practical clinical application value.