High-resolution automatic fault identification method. This is a repository for the paper "What can deep learning-based resolution-improved seismic data do? A case study of faults identification".
An end-to-end workflow for automatic high-resolution fault identification. The original seismic image (a) is fed to the well-trained HRNet to acquire the high-resolution seismic image (b). Then, the well-trained FaultNet obtains the high-resolution fault identification results (c) from the enhanced seismic image.
Fault identification comparison on the Kerry3D seismic survey. (a) Raw seismic image. Fault probability maps (b), (c), and (d) are obtained by feeding (a) directly into FaultNet, FaultSeg3D, and Swin UNETR. (e) Seismic image is generated by HRNet with enhanced resolution and suppressed random noise. (f) High-resolution fault probability map obtained by feeding (e) into FaultNet. Fault annotations of the raw seismic image (g) and the HRNet-enhanced seismic image (h) are made by three interpreters (The red, blue, and green lines indicate the fault lines labeled by different interpreters.). The red arrows in (a) and (e) indicate two adjacent, closely spaced faults, and the yellow arrows indicate faults with small throws.
All training and test code are in the directory FaultSegmentation/code and ImproveResolution/code. And the code for field data application and plotting is in the in the directory Application/Real.
The synthetic seismic data used for training can be obtained by visting the "https://www.kaggle.com/datasets/leilin1995/seisgan".
- python 3.6.13
- pytorch 1.9.1
- torchvision 0.10.1
- tqdm 4.62.3
- scipy 1.5.4
- numpy 1.19.5
- h5py 3.1.0
- pandas 1.1.5
- PIL 8.4.0
- matplotlib 3.3.4
You can use this method by following the example in the application.
If you find this work useful in your research, please consider citing:
Lin, L., Li, C., Kuang, Y.,Xin, X. & Zhong, Z. (2025) Applications of deeplearning-based resolution-enhanced seismic data infault identification. Geophysical Prospecting, 1–20.https://doi.org/10.1111/1365-2478.13664
BibTex