We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
Automatic seismic fault identification based on an improved U-Net network.
- Authors
Wu, Jizhong; Shi, Ying; Wang, Kexin; Yang, Chenyu; Yang, Qianqian
- Abstract
Fault identification is a key step in structural interpretation. Traditional fault identification methods are easily affected by the seismic quality and interpreter experience, and the identification methods of complex fault zones and multiscale faults require further improvement. The U-Net, frequently used at present, has achieved satisfactory results in fault identification. However, this network has a limited ability to extract and recover feature information and does not have a mechanism of concern, limiting its ability to identify complex faults. To solve these problems, this study proposes an improved U-Net network model based on a conventional U-Net network. A multiscale residual module was used to extract the features instead of the U-Net's two-layer convolution. The residual jump connection replaced the U-Net jump connection to avoid semantic loss caused by the fusion of high- and low-level semantic information. An attention mechanism was introduced to integrate the global, local, spatial, and channel features to ensure that the model could extract image features from various dimensions to the maximum extent. An improved U-Net network model was applied to train the model data and test the actual field data. Our results show that the proposed network model is better at fault identification, avoids human interference to a certain extent, and is advantageous compared to the conventional U-Net method.
- Subjects
CONVOLUTIONAL neural networks; FAULT zones; FEATURE extraction; ARTIFICIAL intelligence
- Publication
Acta Geophysica, 2024, Vol 72, Issue 4, p2377
- ISSN
1895-6572
- Publication type
Article
- DOI
10.1007/s11600-023-01200-7