Seismic noise attenuation via convolution neural network with learningnoise prior

The presence of noise degrades the quality of seismic data and makes the subsequent processing tasks and interpretation more challenging. Therefore, seismic noise attenuation is a key step in the processing of seismic data. We propose a novel convolutional neural network (CNN) framework with learning noise prior. Unlike conventional CNN-based seismic denoising methods, this new network is composed of a noise extractor and a denoiser. The noise extractor extracts noise from the original data to provide a high-precision noise prior to the denoising process. The denoiser uses the noise prior for denoising of the seismic data. This method is superior to the presently used networks in terms of the denoising effect. Additionally, the proposed network can be applied for random noise suppression as well as coherent noise attenuation. Synthetic and field tests illustrated the superiority of the proposed approach over the traditional denoising methods in suppressing noise and improving the signal-to-noise ratio (SNR) of seismic data.