Random Noise Attenuation by Self-supervised Learning from Single Seismic Data.

Random noise attenuation is of great importance to obtain high-quality seismic data. Unsupervised deep learning methods have received much attention for various seismic data processing tasks in recent years. Specifically, the self-supervised deep learning method obtains supervisory information from the data itself, showing its promising denoising ability in various geophysical applications. In this work, a dropout-based self-supervised (DSS) deep learning method is applied for single seismic data random noise attenuation. In the DSS method, single masked noisy data is generated for training and self-supervised loss function construction. The U-shaped convolutional network (U-Net) with a dropout strategy is taken as the main network framework to enhance the denoising stability and reduce the over-fitting effectively. Compared with the traditional f-x deconvolution (FX-Decon) and deep image prior (DIP) method, the DSS method achieves better denoising results in preserving details for synthetic seismic data and field data. Moreover, numerical experiments indicate that the DSS method is stable for seismic denoising and reduces the over-fitting phenomenon. [ABSTRACT FROM AUTHOR]