Abstract
Seismic data is contaminated by random noise, which decreases the accuracy of subsequent seismic data processing. Therefore, random noise attenuation is an important step to improve the signal-to-noise ratio (SNR) of seismic data. We design a Deep Image Prior (DIP) network architecture based on U-net for random noise attenuation. However, it is difficult to determine the optimal epoch for the best result during the DIP network training for unlabeled noisy data. Thus, we introduce a new quality control criterion based on adjacent estimations, and a recursive denoising workflow is developed. Compared with the DIP network without the recursive method, the recovered SNR of the proposed recursive denoising is greatly improved. Numerical examples on pre-stack and post-stack synthetic seismic data and a field post-stack data demonstrate the effectiveness of the proposed recursive method based on the DIP.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call