State Key Laboratory for Manufacturing Systems Engineering, Xi'an Jiaotong University, Xi'an 710049, China.
Rev Sci Instrum. 2020 Apr 1;91(4):045104. doi: 10.1063/1.5136269.
At present, denoising parameters in different signal processing algorithms require a specific signal waveform to be set. Human factors would significantly affect the denoising result. To solve this problem, we proposed a signal adaptive denoising method based on a denoising autoencoder to achieve denoising on ultrasonic signals. By applying this method to sample signals and comparing with the singular value decomposition (SVD), principal component analysis (PCA), and wavelet algorithms, it is found that this method can effectively suppress the noise at different noise intensities. Using the signal to noise ratio, root mean square error, and autocorrelation coefficient as evaluation parameters in the experiment, the overall denoising effect of the proposed method is better than that of PCA, and this method is better than the wavelet and SVD algorithms having a relatively weak noise intensity. In addition, by comparing the reconstructed signal curve of the proposed method and that of the wavelet algorithm, the proposed method can retain the information of signal saltation with a better performance. Finally, we apply this method for processing ultrasonic signals and verify its effectiveness from time and frequency domain diagrams.
目前，不同信号处理算法中的去噪参数需要设置特定的信号波形，人为因素会显著影响去噪效果。为解决这一问题，我们提出了一种基于去噪自动编码器的信号自适应去噪方法，以实现对超声波信号的去噪。通过将该方法应用于样本信号，并与奇异值分解（SVD）、主成分分析（PCA）和小波算法进行比较，发现该方法可以有效地抑制不同噪声强度下的噪声。在实验中，使用信噪比、均方根误差和自相关系数作为评价参数，发现该方法的整体去噪效果优于 PCA，并且在噪声强度较弱的情况下，该方法优于小波和 SVD 算法。此外，通过比较所提出方法和小波算法的重建信号曲线，发现所提出方法可以更好地保留信号突变的信息。最后，我们将该方法应用于处理超声信号，并从时间和频域图验证其有效性。
