Seismic PnP

Plug and Play method for post-stack seismic inversion

Welcome to the Seismic PnP project

This project was born during the Machine Learning in Geoscience Course at KAUST by Prof. Matteo Ravasi. It was developed together with my friend and colleague Juan Romero.

Project description

The idea of this work is based on the concept of using Plug-and-Play (PnP) regularization, which suggests reinterpreting the effect of the regularizer as a denoising problem. Due to this denoising step, various statistical and deep denoisers become attractive. For this work, DnCNN (gaussian blind denoiser) and DRUNet (non-blind gaussian denoisers) are used to evaluate if we can show superior results compared to state-of-the-art regularization techniques on post-stack seismic inversion such as TV regularization. It is important to emphasize that we are not denoising the input data; the denoising is performed along the Primal-Dual algorithm .

Status and Results

This new approach presented is able to outperfom the state-of-the art regularization techniques. More detailed information about the work could be found on the publication.

Results obtained with the proposed approach and compared to state-of-the-art inversion procedures.

Publications

More details about the project could be found on the following publication: PnP Seismic.