Ahmad Mustafa, Motaz Alfarraj, and Ghasssan AlRegib
This repository includes the codes for the paper:
'Estimation of Acoustic Impedance from Seismic Data using Temporal Convolutional Network', Expanded Abstracts of the SEG Annual Meeting , San Antonio, TX, Sep. 15-20, 2019. [PDF]
The code has been built in Python 3.7.0 and requires the following dependencies before it can be run:
numpy 1.15.1
matplotlib 3.0.0
pytorch 0.4.1
seaborn 0.9.0
segyio 1.7.0
tensorboard 1.13.1
tensorboardx 1.7
tensorflow 1.13.1
pandas 0.24.2
scipy 1.1.0
In exploration seismology, seismic inversion refers to the process of inferring physical properties of the subsurface from seismic data. Knowledge of physical properties can prove helpful in identifying key structures in the subsurface for hydrocarbon exploration. In this work, we propose a workflow for predicting acoustic impedance (AI) from seismic data using a network architecture based on Temporal Convolutional Network by posing the problem as that of sequence modeling. The proposed workflow overcomes some of the problems that other network architectures usually face, like gradient vanishing in Recurrent Neural Networks, or overfitting in Convolutional Neural Networks. The proposed workflow was used to predict AI on Marmousi 2 dataset with an average r2 coefficient of 91% on a hold-out validation set.
Create a directory called data inside the parent directory containing the project files. Download the data from this
link and unzip the contents of the file in the data folder.
After scrolling to the directory containing the project, in the command line, run:
python train.py
This will run the train.py file with the default settings for all the parameters. To view all these along with their default values, run:
python train.py -h
If you want to change, say the number of epochs to 1200, just run:
python train.py --n_epochs 1200
The training and validation losses will be printed on the terminal along with the epoch number. After the training is over, the images for the ground-truth AI model, the predicted AI model, the absolute difference between the two, and a figure showing hand-picked predicted AI traces superimposed on their ground-truth traces will be printed on the screen. You can also view the training and validation loss curves along with the predicted AI images in tensorboard by running:
tensorboard --logdir path/to/log-directory
If you have found our code and data useful, we humbly request you to cite our work. You can cite the SEG abstract as:
@inbook{doi:10.1190/segam2019-3216840.1,
author = {Ahmad Mustafa and Motaz Alfarraj and Ghassan AlRegib},
title = {Estimation of acoustic impedance from seismic data using temporal convolutional network},
booktitle = {SEG Technical Program Expanded Abstracts 2019},
chapter = {},
pages = {2554-2558},
year = {2019},
doi = {10.1190/segam2019-3216840.1},
URL = {https://library.seg.org/doi/abs/10.1190/segam2019-3216840.1},
eprint = {https://library.seg.org/doi/pdf/10.1190/segam2019-3216840.1},
abstract = { In exploration seismology, seismic inversion refers to the process of inferring physical properties of the subsurface from seismic data. Knowledge of physical properties can prove helpful in identifying key structures in the subsurface for hydrocarbon exploration. In this work, we propose a workflow for predicting acoustic impedance (AI) from seismic data using a network architecture based on Temporal Convolutional Network by posing the problem as that of sequence modeling. The proposed workflow overcomes some of the problems that other network architectures usually face, like gradient vanishing in Recurrent Neural Networks, or overfitting in Convolutional Neural Networks. The proposed workflow was used to predict AI on Marmousi 2 dataset with an average r2 coefficient of 91\% on a hold-out validation set.Presentation Date: Wednesday, September 18, 2019Session Start Time: 9:20 AMPresentation Time: 9:45 AMLocation: Poster Station 1Presentation Type: Poster }
}
The arXiv preprint is available at: https://arxiv.org/abs/1906.02684
The code and the data are provided as is with no guarantees. If you have any questions, regarding the dataset or the code, you can contact me at (amustafa9@gatech.edu), or even better, open an issue in this repo and we will do our best to help.