CNN-POCS algorithm for seismic data interpolation.
This repository contains the reproducible code for the article "Can learning from image denoising be used for seismic data interpolation?" This article can also be reached at Arxiv but it is somehow out of date.
This repository depends on Matlab and matconvnet. Matlab beyond 2018a and matconvnet 1.0beta25 are recommended. CUDA is required for training process/gpu testing. Please refer to Matlab GPU support to setup your environment.
The training code can be found in TrainingCodes.
We provide few demos for reproducing some results. The pre-trained models using natural images are in folder models. The hyperbolic events data and the synthetic 3D data are included in seismicData.
- Demo_pocs_cnn.m is provided for testing the CNN-POCS algorithm for seismic data interpolation.
- Demo_cnndenoise.m is provided for testing denoising 2D seismic data using natural images pretrained CNN models.
- Demo_cnndenoise3D.m is provided for testing denoising 3D seismic data.
If this repository helps you with your research, please consider citing our work
@article{zhang2020can,
title={Can learning from natural image denoising be used for seismic data interpolation?},
author={Zhang, Hao and Yang, Xiuyan and Ma, Jianwei},
journal={Geophysics},
volume={85},
number={4},
pages={1--142},
year={2020},
publisher={Society of Exploration Geophysicists}
}