MIFN

---

The code for paper 'Automatic Velocity Picking Using a Multi-Information Fusion Deep Semantic Segmentation Network' (MIFN)

If you cite this paper, please use the following bibtex:

@article{wang2022automatic,
  title={Automatic Velocity Picking Using a Multi-Information Fusion Deep Semantic Segmentation Network},
  author={Wang, Hongtao and Zhang, Jiangshe and Zhao, Zixiang and Zhang, Chunxia and Long, Li and Yang, Zhiyu and Geng, Weifeng},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  volume={60},
  pages={1--10},
  year={2022},
  publisher={IEEE}
}

0 Preparing

0.1 Python, CUDA and Pytorch

• Python 3.7

• CUDA 11.0

• Pytorch 1.7.1 GPU

  pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html

0.2 Python packages

• install packages based on requirements.txt

pip install -r requirements.txt

0.3 Requirement structure for data folder

• You should follow the structure for repreducing

  --- Root

  ​ |___ segy

  ​ |___ vel.stk.sgy

  ​ |___ vel.pwr.sgy

  ​ |___ vel.gth.sgy

  ​ |___ t_v_labels.dat

1 code structure

MIFN-VELOCITY-PICKING
│  predict.py                 # main predict 
│  README-ch.md               # readme in CH
│  README.md                  # readme in EN
│  requirements.txt           # python packages list
│  test.py                    # main test
│  train.py                   # main train
│
├─loss  // loss function
│      detail_loss.py
│      loss.py
│      util.py
│      __init__.py
│
├─net  // our proposed MIFN
│      AblationNet.py
│      BasicModule.py
│      MIFNet.py
│      __init__.py
│
├─Tuning  // tuning
│      tuning.py
│
└─utils  // other tools
        BuiltStkDataSet.py
        evaluate.py
        GetNMOResult.py
        LabTxt2Npy.py
        LoadData.py
        logger.py
        metrics.py
        PastProcess.py
        PlotTools.py
        remove.py
        SpecEnhanced.py
        __init__.py

2 train your model

Please run the following code in the repo root.

2.1 transfer labels file type and generate h5 datasets

# Transfer label file "t_v_labels.dat" to “t_v_labels.npy”
python utils/LabTxt2Npy.py /Root/t_v_labels.dat /Root/t_v_labels.npy
# Make H5 dataset
python utils/BuiltStkDataSet.py /Root

2.2 training processing

# run on cmd (windows) or terminal (linux)
python train.py --DataSetRoot Root --DataSet SetName --GatherLen GatherLen --SeedRate 0.6 --trainBS 16
# an example
python train.py --DataSetRoot /home/colin/data/Spectrum/hade --DataSet hade --GatherLen 15 --SeedRate 0.6 --trainBS 16

3 predict

# run on cmd (windows) or terminal (linux)
python predict.py --LoadModel PthPath --DataSetRoot Root --DataSet SetName --GatherLen GatherLen --PredBS 16

# an example
python predict.py --LoadModel /home/colin/Project/spectrum/MIFN-Submit/result/hade/model/hade_256_0.6.pth --DataSetRoot /home/colin/data/Spectrum/hade --DataSet hade --GatherLen 15 --PredBS 16

tips：results saved in CSV file, including four columns: 1 line, 2 trace, 3 time, 4 velocities.