A synthetic denoising algorithm for full-waveform induced polarization based on deep learning

• DOI: 10.1190/geo2022-0234.1
• Publication Date: 2022-10-31T17:09:30Z
• Authors: Weiqiang Liu; Qingtian Lü; Rujun Chen; Pinrong Lin; Jiayong Yan; Kun Zhang; Regean Pumulo Pitiya
• Keywords: Autoencoder; Smoothing; SIGNAL (programming language)

Induced polarization (IP) is a widely used geophysical exploration technique. Continuous random noise one of the most prevalent interferences that can seriously contaminate IP signal and distort apparent electrical characteristics. We develop separation algorithm based on deep learning to overcome this issue. The standard signals are first produced by combining Cole-Cole model Fourier series decomposition, then mathematical simulation generate various types interferences, which subsequently added signals. Then, denoising autoencoder neural network structure built trained using noisy as input samples pure output samples. resulting optimum capable automatically reconstructing clean from input. This tested synthetic data sets. perform reduction thousands survey points in matter seconds reduce distortion approximately 25% less than 5%. Deep learning-based provides superior computation speed precision compared with wavelet smoothing filtering approach. for high-quality do not vary considerably before after reduction. successfully suppressed low-quality Based findings, has promising future suppressing aid improving quality high efficiency precision.