*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: Large Scale Wavefield Reconstruction via Weighted Matrix Factorization
Committee:
Dr. Herrman, Advisor
Dr. AlRegib, Chair
Dr. Peng
Abstract: The objective of the proposed research is to improve the performance of wavefield recovery at the high frequencies and to increase its efficiency through the use of a parallel algorithm. Seismic data acquisition is a critical phase in the early stages of oil and gas exploration. Inspired by relatively recent developments encouraged by the field of Compressive Sensing(CS), seismic data is increasingly collected randomly along the spatial coordinates to shorten the acquisition time and to reduce cost. While random sampling increases acquisition efficiency, it shifts the burden from field acquisition to data processing, as fully sampled seismic data is required for subsequent steps such as multiple removal and migration. Rank minimization is an effective way to recovering the missing trace data. Unfortunately, this technique’s performance degrades with increasing frequency, as high-frequency data are not always accurately captured by low-rank matrix factorizations. As a result, recovered data often suffers from low signal to noise ratio (SNR)s at the higher frequencies. To deal with this situation, we propose a recursively weighted recovery method that improves the performance at the high frequencies by recursively using information from matrix factorizations at neighboring lower frequencies. To expand the application of the new method to 3D seismic acquisitions, we proposed a parallelized alternating optimization approach for parallelizing the weighted low-rank factorization algorithm. By working with eight parallel workers (two threads each) in the Cloud, we are able to achieve a significantly faster runtime (83 ) than the original weighted method while maintaining the SNRs that is very close to the result obtained with the original formulation. Even though this new approach has had success, there remains the challenge that land seismic data contains ground roll, which because of its strong amplitude and high spatial frequency content is known to degrade the wavefield reconstructions based on promoting structure whether this is sparsity or low rank. I would like to propose a practical workflow for reducing the noise introduced by ground roll and improving the land seismic wavefield recovery with more reflection and diffraction information by using parallel weighted matrix factorization.
