Visual Explanations from Convolutional Neural Networks for Fault Detection

Zhining Liu,Chengyun Song,Bin She,Kunhong Li,Xingmiao Yao,Guangmin Hu

DOI: https://doi.org/10.1190/segam2018-2997882.1

2018-01-01

Abstract:PreviousNext No AccessSEG Technical Program Expanded Abstracts 2018Visual explanations from convolutional neural networks for fault detectionAuthors: Zhining LiuChengyun SongBin SheKunhong LiXingmiao YaoGuangmin HuZhining LiuChongqing University of TechnologySearch for more papers by this author, Chengyun SongChongqing University of TechnologySearch for more papers by this author, Bin SheUniversity of Electronic Science and Technology of ChinaSearch for more papers by this author, Kunhong LiUniversity of Electronic Science and Technology of ChinaSearch for more papers by this author, Xingmiao YaoUniversity of Electronic Science and Technology of ChinaSearch for more papers by this author, and Guangmin HuUniversity of Electronic Science and Technology of ChinaSearch for more papers by this authorhttps://doi.org/10.1190/segam2018-2997882.1 SectionsAboutPDF/ePub ToolsAdd to favoritesDownload CitationsTrack CitationsPermissions ShareFacebookTwitterLinked InRedditEmail AbstractDeep learning has emerged as the leading technology in various disciplines for its superior performance. Recent research demonstrates that deep learning would be successfully applied to process seismic data. Even though deep neural networks enable superior performance, their lack of intuitive and understandable explanations makes this technique become less trustworthy. In this abstract, we introduce a technique, Smooth-Grad, to give visual explanations from a convolutional neural network (CNN) that is trained for fault detection. This technique highlights the regions of an input that are particularly influential to the final classification, which is often called the sensitivity map. By analyzing the sensitivity map generated by the CNN trained for fault detection, we find that the CNN does learn some features that are effective for fault detection, but the way a CNN gives its interpretation is still far from human interpretation.Presentation Date: Tuesday, October 16, 2018Start Time: 1:50:00 PMLocation: Poster Station 2Presentation Type: PosterKeywords: neural networks, machine learning, faults, interpretationPermalink: https://doi.org/10.1190/segam2018-2997882.1FiguresReferencesRelatedDetailsCited byInterpretability-Guided Convolutional Neural Networks for Seismic Fault Segmentation SEG Technical Program Expanded Abstracts 2018ISSN (print):1052-3812 ISSN (online):1949-4645Copyright: 2018 Pages: 5520 publication data© 2018 Published in electronic format with permission by the Society of Exploration GeophysicistsPublisher:Society of Exploration Geophysicists HistoryPublished Online: 27 Aug 2018 CITATION INFORMATION Zhining Liu, Chengyun Song, Bin She, Kunhong Li, Xingmiao Yao, and Guangmin Hu, (2018), "Visual explanations from convolutional neural networks for fault detection," SEG Technical Program Expanded Abstracts : 2226-2230. https://doi.org/10.1190/segam2018-2997882.1 Plain-Language Summary Keywordsneural networksmachine learningfaultsinterpretationPDF DownloadLoading ...