Digital Talk

Abstract – Deep Learning for Seismic Applications

Seismic imaging is an essential step of hydrocarbon exploration and reservoir characterization. Unfortunately, recorded seismic signals at the surface are inevitably contaminated by coherent and incoherent noise of various nature. The process of removing the noise, while retaining the primary signal, is called seismic processing, and it is usually a challenging and tedious task. In this talk, we focus on the relevant, well-known seismic processing task of multiple removal. This mainly consists of getting rid of coherent noise caused by reverberations of waves between reflectors. To achieve that, we employ diffusion probabilistic models. This novel class of generative models has, in a short period of time, achieved seminal results in various image-processing tasks such as editing, interpolation, denoising, and super-resolution.