The Separation of Seismic Surface-Related Multiples Based on CAE-SAGAN

Tao Liu-Rong,Jiang Jin-Sheng,Ren Hao-Ran,Ye Yue-Ming,Wu Bang-Yu
DOI: https://doi.org/10.1190/segam2021-3590505.1
2021-01-01
Abstract:PreviousNext No AccessFirst International Meeting for Applied Geoscience & Energy Expanded AbstractsThe separation of seismic surface-related multiples based on CAE-SAGANAuthors: Tao Liu-RongJiang Jin-ShengRen Hao-RanYe Yue-MingWu Bang-YuTao Liu-RongSchool of Earth Sciences, Zhejiang UniversitySearch for more papers by this author, Jiang Jin-ShengSchool of Earth Sciences, Zhejiang UniversitySearch for more papers by this author, Ren Hao-RanSchool of Earth Sciences, Zhejiang UniversitySearch for more papers by this author, Ye Yue-MingPetro China Hang Zhou Research Institute of GeologySearch for more papers by this author, and Wu Bang-YuSchool of Mathematics and Statistics, Xi’an Jiaotong UniversitySearch for more papers by this authorhttps://doi.org/10.1190/segam2021-3590505.1 SectionsSupplemental MaterialAboutPDF/ePub ToolsAdd to favoritesDownload CitationsTrack CitationsPermissions ShareFacebookTwitterLinked InRedditEmail AbstractThe multiple separation technology in data space can avoid the dependence of prior geological structure information and the expensive computation cost. Deep neural networks can automatically mine features from data. Once the network is successfully trained, it has extremely high data processing efficiency. In this study, a Generative Adversarial Network (GAN) framework is proposed to separate primaries from surface-related multiples. Here, the generator is constructed by a Convolutional Auto-Encoder (CAE) network, and the discriminator is constructed by a convolutional neural network. In order to increase the receptive field of network, we have also introduced a Self-Attention (SA) mechanism. Therefore, this network is called CAE-SAGAN. We use finite difference forward modeling to generate a dataset for training the network, including data that mixes primaries with surface-related multiples, and data that contains only primaries. These two kinds of data are applied to the input and label datasets of the network respectively. The test dataset consists of synthetic data that did not participate in training. Experiments in the data domain and the results of reverse time migration (RTM) imaging show that this method can separate surface-related multiples automatically and efficiently.Keywords: multiples, machine learning, reverse time migrationPermalink: https://doi.org/10.1190/segam2021-3590505.1FiguresReferencesRelatedDetails First International Meeting for Applied Geoscience & Energy Expanded AbstractsISSN (print):1052-3812 ISSN (online):1949-4645Copyright: 2021 Pages: 3561 publication data© 2021 Published in electronic format with permission by the Society of Exploration GeophysicistsPublisher:Society of Exploration Geophysicists HistoryPublished: 01 Sep 2021 CITATION INFORMATION Tao Liu-Rong, Jiang Jin-Sheng, Ren Hao-Ran, Ye Yue-Ming, and Wu Bang-Yu, (2021), "The separation of seismic surface-related multiples based on CAE-SAGAN," SEG Technical Program Expanded Abstracts : 2914-2918. https://doi.org/10.1190/segam2021-3590505.1 Plain-Language Summary Keywordsmultiplesmachine learningreverse time migrationPDF DownloadLoading ...
What problem does this paper attempt to address?