Seismic Source Recovery Algorithm via Internal Learning in the Cross-spread Domain: A Tensorflow Implementation

This is a Tensorflow implementation of the proposed work in "Seismic Source Recovery Algorithm via Internal Learning in the Cross-spread Domain" (Sebastián Rivera, Iván Ortíz, Tatiana Gelvez, Laura Galvis, Henry Arguello, EAGE, 2022).

Project structure

.
├── ...
├── data                   # Folder to place data to reconstruct.
│   └── cube.npy            # Tridimensional array (HxWxC).
├── src                    # Source code.
│   ├── config.yml          # Hyperparameters to change.
│   ├── default.py          # Default hyperparameters.
│   ├── metrics.py          # Metrics to measure the performance.
│   ├── models.py           # Neural network architecture.
│   ├── preprocessing.py    # Preprocessing operations before the internal learning.
│   ├── utils.py            # Utils functions.
│   └── main.py             # Internal learning training.
├── docker-compose.yml
├── Dockerfile
└── requirements.txt

Usage

To avoid Tensorflow and CUDA compatibility issues, we employ and recommend Docker. After installing Docker, execute the following command in the project's root folder:

docker-compose up 

Results

The image below illustrates a comparison between the reconstructed shots obtained using the proposed method and those generated by other reconstruction techniques. The top-right corner displays the PSNR metric.

Cite

@inproceedings{rivera2022seismic,
  title={Seismic Source Recovery Algorithm via Internal Learning in the Cross-spread Domain},
  author={Rivera, S and Ortiz, I and Gelvez-Barrera, T and Galvis, L and Arguello, H},
  booktitle={Fourth HGS/EAGE Conference on Latin America},
  volume={2022},
  number={1},
  pages={1--5},
  year={2022},
  organization={European Association of Geoscientists \& Engineers}
}

License