Reconstruction of seismic field data with convolutional U-Net considering the optimal training input data
- Authors
- Park, Jiho; Yoon, Daeung; Seol, Soon Jee; Byun, Joongmoo
- Issue Date
- Sep-2020
- Publisher
- Society of Exploration Geophysicists
- Citation
- SEG International Exposition and Annual Meeting 2019, pp.4650 - 4654
- Indexed
- SCOPUS
- Journal Title
- SEG International Exposition and Annual Meeting 2019
- Start Page
- 4650
- End Page
- 4654
- URI
- https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/145130
- DOI
- 10.1190/segam2019-3216017.1
- Abstract
- Deep learning (DL) methods are recently used as a powerful tool in seismic signal processing. Most of seismic trace reconstruction methods are governed by the super-resolution methods based on the convolutional neural network (CNN). The performances of these kinds of methods depend on not only how training model is constructed but also what is learned from training data, especially on field data application. In this study, we propose two sequences of seismic trace interpolation through t-SNE and convolutional U-Net to provide a guide to the optimal organization of training sets and to successful reconstruction of missing seismic traces. We test the proposed method on the Ocean Bottom Cable (OBC) field data to evaluate performances. The common receiver gather (CRG) as well as another common shot gather (CSG) are reasonably interpolated by the convolutional U-Net model trained with mixed data sets from two CSGs and a spatial aliasing is also reduced properly.
- Files in This Item
-
Go to Link
- Appears in
Collections - 서울 공과대학 > 서울 자원환경공학과 > 1. Journal Articles
![qrcode](https://api.qrserver.com/v1/create-qr-code/?size=55x55&data=https://scholarworks.bwise.kr/hanyang/handle/2021.sw.hanyang/145130)
Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.