2020
DOI: 10.48550/arxiv.2002.00031
|View full text |Cite
Preprint
|
Sign up to set email alerts
|

Optimal Transport Based Seismic Inversion: Beyond Cycle Skipping

Abstract: Full waveform inversion (FWI) is today a standard process for the inverse problem of seismic imaging. PDE-constrained optimization is used to determine unknown parameters in a wave equation that represent geophysical properties. The objective function measures the misfit between the observed data and the calculated synthetic data, and it has traditionally been the least-squares norm. In a sequence of papers, we introduced the Wasserstein metric from optimal transport as an alternative misfit function for mitig… Show more

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1

Citation Types

0
6
0

Year Published

2020
2020
2023
2023

Publication Types

Select...
3
2

Relationship

2
3

Authors

Journals

citations
Cited by 5 publications
(6 citation statements)
references
References 71 publications
0
6
0
Order By: Relevance
“…Typically, the zero-gradient condition is far from enough to guarantee the optimality, especially for FWI. There has been extensive literature on tackling the nonconvexity [9,31]. In this paper, we focus on accelerating the convergence and not addressing the cycle-skipping issues, which is another important research topic by itself.…”
Section: Methodsmentioning
confidence: 99%
“…Typically, the zero-gradient condition is far from enough to guarantee the optimality, especially for FWI. There has been extensive literature on tackling the nonconvexity [9,31]. In this paper, we focus on accelerating the convergence and not addressing the cycle-skipping issues, which is another important research topic by itself.…”
Section: Methodsmentioning
confidence: 99%
“…However, the exponential normalization method cannot keep the amplitude ratio with the amplified positive and suppressed negative parts of the seismic data. Especially for data with low SNR in practical applications, the exponential scaling method has a risk of amplifying noise (Engquist & Yang, 2020). Thus, the phase information cannot be extracted exactly.…”
Section: Data Normalization and Gradient Calculationmentioning
confidence: 99%
“…In contrast, the affine scaling strategy is implemented by adding a constant value to the signals. Although it works well for practical applications, this method cannot provide a strict convex misfit function with a relatively wider convergence range compared to the conventional L 2 objec-tive function (Engquist & Yang, 2020). As a common normalization method, an exponentiation strategy can provide good convexity by setting appropriate parameters.…”
Section: Introductionmentioning
confidence: 99%
“…The Wasserstein metric has been actively studied since the late 20th century [65]. Recently, it is proved to have unique features as a loss function in numerous applied fields, such as image processing [33], machine learning [6], large-scale inverse problems [21] and statistical inference [9], only to mention a few. The research development of optimal transport in applied mathematics is so rapid that we do not attempt to give a comprehensive overview.…”
mentioning
confidence: 99%
“…Second, the basin of attraction for these chaotic dynamics is compactly supported on a low-dimensional subset of R d . The Wasserstein metric is well-defined for densities of different support as it not only considers the local intensity differences but also incorporates the global geometry mismatches [21], which is not the case of Kullback-Leibler divergence. The L p norm also suffers if the supports are disjoint and may produce wrong or random update directions if supports only partially intersect.…”
mentioning
confidence: 99%