Low-frequency extrapolation with deep learning
The lack of the low frequency information and good initial model can seriously affect the success of full waveform inversion (FWI) due to the inherent cycle skipping problem. Reasonable and reliable low frequency extrapolation is in principle the most direct way to solve this problem. In this paper,...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
Society of Exploration Geophysicists
2020
|
| Online Access: | https://hdl.handle.net/1721.1/124443 |
| _version_ | 1826198894758854656 |
|---|---|
| author | Sun, Hongyu Demanet, Laurent |
| author2 | Massachusetts Institute of Technology. Department of Earth, Atmospheric, and Planetary Sciences |
| author_facet | Massachusetts Institute of Technology. Department of Earth, Atmospheric, and Planetary Sciences Sun, Hongyu Demanet, Laurent |
| author_sort | Sun, Hongyu |
| collection | MIT |
| description | The lack of the low frequency information and good initial model can seriously affect the success of full waveform inversion (FWI) due to the inherent cycle skipping problem. Reasonable and reliable low frequency extrapolation is in principle the most direct way to solve this problem. In this paper, we propose a deep-learning-based bandwidth extension method by considering low frequency extrapolation as a regression problem. The Deep Neural Networks (DNNs) are trained to automatically extrapolate the low frequencies without prepro-cessing steps. The band-limited recordings are the inputs of the DNNs and, in our numerical experiments, the pretrained neural networks can predict the continuous-valued seismograms in the unobserved low frequency band. For the numerical experiments considered here, it is possible to find the amplitude and phase correlations among different frequency components by training the DNNs with enough data samples, and extrapolate the low frequencies from the band-limited seismic records trace by trace. The synthetic example shows that our approach is not subject to the structural limitations of other methods to bandwidth extension, and seems to offer a tantalizing solution to the problem of properly initializing FWI. |
| first_indexed | 2024-09-23T11:11:38Z |
| format | Article |
| id | mit-1721.1/124443 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T11:11:38Z |
| publishDate | 2020 |
| publisher | Society of Exploration Geophysicists |
| record_format | dspace |
| spelling | mit-1721.1/1244432022-10-01T01:57:08Z Low-frequency extrapolation with deep learning Sun, Hongyu Demanet, Laurent Massachusetts Institute of Technology. Department of Earth, Atmospheric, and Planetary Sciences Massachusetts Institute of Technology. Department of Mathematics The lack of the low frequency information and good initial model can seriously affect the success of full waveform inversion (FWI) due to the inherent cycle skipping problem. Reasonable and reliable low frequency extrapolation is in principle the most direct way to solve this problem. In this paper, we propose a deep-learning-based bandwidth extension method by considering low frequency extrapolation as a regression problem. The Deep Neural Networks (DNNs) are trained to automatically extrapolate the low frequencies without prepro-cessing steps. The band-limited recordings are the inputs of the DNNs and, in our numerical experiments, the pretrained neural networks can predict the continuous-valued seismograms in the unobserved low frequency band. For the numerical experiments considered here, it is possible to find the amplitude and phase correlations among different frequency components by training the DNNs with enough data samples, and extrapolate the low frequencies from the band-limited seismic records trace by trace. The synthetic example shows that our approach is not subject to the structural limitations of other methods to bandwidth extension, and seems to offer a tantalizing solution to the problem of properly initializing FWI. United States. Air Force. Office of Scientific Research (Grant FA9550-17-1-0316) National Science Foundation (U.S.) (Grant DMS1255203) 2020-03-31T13:37:59Z 2020-03-31T13:37:59Z 2018-08 2019-11-12T14:02:58Z Article http://purl.org/eprint/type/ConferencePaper 1949-4645 1052-3812 https://hdl.handle.net/1721.1/124443 Sun, Hongyu and Demanet, Laurent. "Low-frequency extrapolation with deep learning." SEG Technical Program Expanded Abstracts (2018): 2011-2015 © 2018 Author(s) en 10.1190/segam2018-2997928.1 SEG Technical Program Expanded Abstracts 2018 Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf Society of Exploration Geophysicists MIT web domain |
| spellingShingle | Sun, Hongyu Demanet, Laurent Low-frequency extrapolation with deep learning |
| title | Low-frequency extrapolation with deep learning |
| title_full | Low-frequency extrapolation with deep learning |
| title_fullStr | Low-frequency extrapolation with deep learning |
| title_full_unstemmed | Low-frequency extrapolation with deep learning |
| title_short | Low-frequency extrapolation with deep learning |
| title_sort | low frequency extrapolation with deep learning |
| url | https://hdl.handle.net/1721.1/124443 |
| work_keys_str_mv | AT sunhongyu lowfrequencyextrapolationwithdeeplearning AT demanetlaurent lowfrequencyextrapolationwithdeeplearning |