Classification and Suppression of Blending Noise Using Convolutional Neural Networks

Abstract

Over the last few years, machine learning has become more and more a topic of interest in the seismic industry. In seismic interpretation like fault/salt dome detection (Amin et al. 2015, Guitton et al. 2017) and velocity picking (Smith 2017), there already have been successful implementations for some years now. Recently, machine learning was introduced in seismic processing algorithms like denoising, regularization and tomography (Araya-Polo et al. 2018) as well. In this abstract a deblending algorithm is proposed that utilizes supervised machine learning algorithms. The method combines the two main functionalities of supervised learning, classification and regression to achieve maximum control on the deblending process. First, blended acquisition and conventional deblending methods are discussed, followed by an introduction to machine learning algorithms and how these machine learning methods can contribute to improve existing deblending algorithms. Finally, synthetic data examples are shown to illustrate the machine learning deblending approach.