Automatic identification of seismic faults via integrating Residual Network-50 residual blocks and convolutional block attention modules

• Published in: Applied geophysics 2023-03, Vol.20 (1), p.20-35
• Main Authors: Wang, Xin-Wei, Shi, Su-Zhen, Yao, Xu-Jun, Pei, Jin-Bo, Wang, Yi-Fan, Yang, Han-Bo, Liu, Dan-Qing
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Data augmentation; Earth and Environmental Science; Earth Sciences; Fault detection; Fault lines; Geophysics/Geodesy; Geotechnical Engineering & Applied Earth Sciences; Human error; Identification; Identification methods; Machine learning; Modules; Neural networks; Seismic data; Seismic Interpretation; Seismological data
• ISSN: 1672-7975; 1993-0658
• DOI: 10.1007/s11770-023-1014-2

Traditional fault identification involves manual marking by geological interpreters, which is time consuming, inefficient, and prone to human error. To address these issues and increase the accuracy of fault identification, a deep-learning-based fault identification method is proposed that uses an attention mechanism to focus on target features. A convolutional block attention module (CBAM) is used in the decoding layer of the U-Net network, and a ResNet-50 residual block is used in the encoding layer. Consequently, a fault identification method based on convolutional neural networks is established and referred to as Res-CBAM-UNet. To enhance the generalization ability of the network model, data augmentation on synthetic seismic data and their corresponding fault labels was performed, and the model was trained using the newly generated training dataset as the input. Subsequently, the model was compared and analyzed with CBAM-UNet, ResNet34-UNet, and ResNet50-UNet networks and tested using the seismic data from actual working areas. Results reveal that the designed Res-CBAM-UNet network has good fault identification performance with high continuity of identified faults and computational efficiency.