Machine learning for seismic processing: The path to fulfilling promises

Machine learning (ML) has garnered great attention within the field of seismic processing due to its vast achievements for quality and efficiency in the area of computer vision. Recent academic papers have demonstrated some potential for the use of machine learning in processing seismic signal, such as random and coherent noise removal, deblending, and interpolation. In this paper, we illustrate some uses of ML on real 3D seismic data and discuss the common challenges that need to be addressed in order to fulfill the promises of the deep neural network (DNN) for seismic processing. We also point out that, in some cases, the result of ML could be good enough for some fit-for-purpose applications. Finally, we summarize a few learnings based on our research and experiences in both the seismic processing and ML worlds.