DNNs in Automatic Salt Identification: How Effective Are They, and How Do We Rank their Performance?

Abstract

Summary Several areas of Earth with large accumulations of oil and gas also have huge deposits of salt below the surface and identifying where large salt deposits are precisely is very difficult. Currently seismic imaging still requires expert human interpretation of salt bodies. This leads to very subjective, highly variable renderings therefore to potentially dangerous situations for oil and gas company drillers. Deep learning algorithms have been used to solve several subsurface imaging tasks such as classification and segmentation. These algorithms are part of the concept of automatic seismic interpretation (ASI), which is now enabling seismic interpreters to complete routine interpretation tasks much more efficiently than what could be done using legacy software. So, how do these ASI networks work when the salt identification task is considered? How efficient are they? What is the computational cost? How good are their outputs? How can we measure their performance and the value they add?