Deep Learning Swell Noise Estimation

Abstract

Towed-streamer data are often contaminated by low-frequency, high-amplitude noise caused by the streamer motion, cable depth controllers, or sea surface waves. So-called swell-noise usually affects a number of neighbouring traces and appears on the shot record as vertical striping. Attenuating this strong noise over a weak reflection signal can be a significant challenge. In this work, we describe a deep learning approach for estimating and subtracting such noise from the recorded data. Inspired by the ideas of the residual network and the generative adversarial network, we have developed a conditional generative adversarial network to estimate swell noise which could also be applicable to other types of noise found in seismic data. We demonstrate the effectiveness of the proposed network by estimating a high-quality swell noise model on a field data example.