Seismic Data Interpolation using Convolutional Neural Network

Prerequisites

• Linux
• Python 3
• Nvidia GPU + CUDA cuDNN

Requirements

• Pytorch >= 0.4
• opencv-python
• numpy

Getting Started

Installation

• Install PyTorch and other dependencies

pip install -r requirements.txt

• Clone this repo:

git clone https://github.com/YinengXiong/SeisInterp.git
cd SeisInterp

Datasets

Download dataset and place it in ./SeisInterp/Data/
The dataset has 96 train, 12 validation and 12 testing shot record, respectively. Each shot record has the size of 500 X 6001 (500 traces with 10m interval, 6001 time sampling points with 0.5 ms)

Google Drive | Baidu Downloads Code: q97i

SeisInterp Train/Test

For example, if you want to train an interpolation model which needs pre-interpolation and only interpolate on X- direction at the scale of 4

python train.py --gpu 0 --dataroot ./Data/MarmousiP20HzAGC500/ --num_traces 500 --nComp 1 --prefix shotp --scale 4 --diretion 0 --arch vdsr

Citation

If you use this code for your research, please cite our papers.

@inproceedings{xiong2019efficient,
  title={Efficient Seismic Data Interpolation Using Deep Convolutional Networks and Transfer Learning},
  author={Xiong, Y and Cheng, J},
  booktitle={81st EAGE Conference and Exhibition 2019},
  volume={2019},
  number={1},
  pages={1--5},
  year={2019},
  organization={European Association of Geoscientists \& Engineers}
}