Several rank-reduction techniques have been proposed to simultaneously denoise and reconstruct seismic datasets. We reformulate the Singular Spectrum Analysis (SSA) filter as a convex optimization problem constraining the associated Hankel matrix to be of low-rank. The Hankel matrix is written as the product of two matrices of lower dimension, which are obtained using a gradient descent algorithm, called the bifactored gradient descent (BFGD). The BFGD is an efficient nonconvex method which can be easily adaptable to include sampling operators within robust measures as cost functions, thus simultaneously handling missing traces and erratic noise. We evaluate the BFGD-based SSA in the simultaneous reconstruction and denoising of a 3D field dataset and compare it with the MSSA interpolation method. The results support that the BFGD does have a competitive performance for seismic data processing applications.
Presentation Date: Monday, September 16, 2019
Session Start Time: 1:50 PM
Presentation Start Time: 3:05 PM
Location: Poster Station 13
Presentation Type: Poster
