This GitHub repository accompanies the article titled: "Efficient extraction of seismic horizons with Deep Learning," submitted to Computers & Geosciences.
We propose a procedure for the interpretation of horizons in seismic reflection data using a Neural Network (NN) approach that is fast, accurate, and reduces the intrinsic subjectivity of manual or control-point-based methods. The training employs a Long Short Term Memory (LSTM) architecture and is performed on synthetic data generated from a convolutional model-based scheme. The extraction step can be applied to any type of field seismic dataset. To enhance the NN's performance across various field conditions, synthetic data are contaminated with different types of noise. The procedure has been tested successfully on both 2-D and 3-D synthetic and field seismic datasets, as well as on Ground Penetrating Radar (GPR) data, demonstrating its versatility and potential. The algorithm operates on a fully 1-D basis and does not require any interpreter input, as necessary thresholds are automatically estimated. Additionally, the prediction includes an associated probability, providing an automatic quantification of the results' reliability.
In this repository, you will find the Jupyter notebooks containing the code used, the trained model, and the data utilized in the article.
model_final.h5: The trained deep learning model used for horizon extraction from seismic data.notebooks/:01_data_gen.ipynb: Notebook for generating synthetic data used in training.02_NN_training.ipynb: Notebook for training the neural network on the generated synthetic data.03_prediction.ipynb: Notebook for applying the trained model to predict horizons on new data.
data/: Contains sample datasets, including synthetic and field seismic data used in the study.scripts/: Python scripts used for data preprocessing, model training, and prediction.
To use the trained model (model_final.h5) for horizon extraction on new data, follow the instructions in the 03_prediction.ipynb notebook. This notebook provides step-by-step guidance on loading the model, preprocessing the input data, and generating horizon probability maps.
The interpretation of the horizon probability maps generated by the model can be performed by integrating the results with other geophysical data. This process is demonstrated in the 03_prediction.ipynb notebook, where examples from the study are provided.
The repository includes the following key results:
- Horizon probability maps showing the likelihood of subsurface reflectors.
- Integrated interpretations combining radar data with seismic datasets to contextualize subsurface structures within the geological framework.
We welcome contributions to this repository. If you have suggestions for improvements or additional features, please submit a pull request or open an issue.
This project is licensed under the MIT License. See the LICENSE file for details.
If you use this code or data in your research, please cite the original paper:
G. Roncoroni, E. Forte, L. Bortolussi, M. Pipan. Efficient extraction of seismic horizons with Deep Learning, Computers & Geosciences, Volume 166, 2022, ISSN 0098-3004, https://doi.org/10.1016/j.cageo.2022.105190