Home

Quick Links

Search

 
Seismic Image Denoising Using Convolutional Neural Network with Residual Learning ApproachNormal access

Authors: S. Wu and Y. Wang
Event name: 81st EAGE Conference and Exhibition 2019
Session: Noise Attenuation II
Publication date: 03 June 2019
DOI: 10.3997/2214-4609.201900851
Organisations: EAGE
Language: English
Info: Extended abstract, PDF ( 1.17Mb )
Price: € 20

Summary:
Random noise attenuation always played an important role in seismic data processing. This study introduces an effective deep learning approach for seismic noise attenuation. The method design a deep feed forward denoising convolutional neural networks with residual learning approach. It learns the noise from the noisy images instead of the latent clean images and obtains the denoised images by subtracting the learned residual from the noisy image. Moreover, the new representative achievements integrated with the residual learning include rectified linear unit and batch normalization. Then, we train the CNN model with poststack field datasets and use the model to suppress the random noise. The results of the field data reveal that the algorithm can remove the random noise and highlight the locally continuous reflectors without losing the resolution of these features.


Back to the article list