No Detail Found!

Denoising seismic signal using a s transform based CNN

Author(s)
LINLANG ZHOU Chuanzhong oil and gas mines of PetroChina Southwest Oil & gas field company, Suining, Sichuan, China
ZIKE MA Chuanzhong oil and gas mines of PetroChina Southwest Oil & gas field company, Suining, Sichuan, China
LIMEI REN Chuanzhong oil and gas mines of PetroChina Southwest Oil & gas field company, Suining, Sichuan, China
JIANDONG YANG Chuanzhong oil and gas mines of PetroChina Southwest Oil & gas field company, Suining, Sichuan, China
LONGYAN YI Chuanzhong oil and gas mines of PetroChina Southwest Oil & gas field company, Suining, Sichuan, China
QING ZHAO Chuanzhong oil and gas mines of PetroChina Southwest Oil & gas field company, Suining, Sichuan, China
JIACHUN YOU School of Geophysics, Chengdu University of technology, Chengdu, Sichuan, China
Abstract
In the field of seismic data processing, signal denoising is an unavoidable and emerging topic. Effectively denoising of seismic signals is a necessary prerequisite for seismic data interpretation and processing. With the development of deep neural networks, a convolution neural network (CNN) denoising method based on continuous S transform (S-transform) is adopted, which is different with the denoising approaching in time domain. In numerical experiments, we denoise the seismic signals with different degrees of noises, by transforming signal within the time domain into the time-frequency map through S-transform method. Then we use CNN for the denoising processing and compare the time-frequency image based on the energy spectrum with the complex spectrum. We have compared the performance of our proposed denoising method with traditional filtering methods including a low-pass denoising filtering, wavelet denoising filtering and median denoising filtering. The three indexes, PSNR, SSIM and RMSE, are used to evaluate the denoising performances, denoising results shows the superiority and effectiveness of our proposed CNN method. In order to improve the performances of signal denoising (with other features) by using our proposed method, we introduce transfer learning to fine-tune our pretrained model. The application of transfer learning shows that our method can be applicable for practical signal processing.
Publication Details
Page(s) 120-128
DOI DOI not available
Published Journal: Geological Bulletin of the Punjab University, Volume: 46, Issue: 0, Year: 2022

Request the full text from PASTIC

Keywords
Denoising seismic signal, Transform based CNN
More from Geological Bulletin of the Punjab University
0

Citations

0

Downloads

24

Views

Copyright ©  PASTIC National Center, Islamabad www.pastic.gov.pk