Deep Learning from Simulated Data for Flood and Debris Flow Mapping

Code for the paper Breaking the Limits of Remote Sensing by Simulation and Deep Learning for Flood and Debris Flow Mapping.

Installation

If necessary create a new Python environment. Install requirements.

pip install -r requirements.txt

Dataset Preparation

The datasets presented in the paper (Northern Kyushu 2017 and Western Japan 2018) can be download here. Please unzip the file in the directory data.

Training and Testing

# To train on the Northern Kyushu 2017 dataset for flood mapping, for example.
python dlsim.py --data NK2017 --train_test train --type wl

# To test pretrained models
python dlsim.py --data NK2017 --train_test test --type wl

Citation

@article{yokoya2020dlsim,
  title={Breaking the Limits of Remote Sensing by Simulation and Deep Learning for Flood and Debris Flow Mapping},
  author={Yokoya, Naoto and Yamanoi, Kazuki and He, Wei and Baier, Gerald and Adriano, Bruno and Miura, Hiroyuki and Oishi, Satoru},
  journal={arXiv:2006.05180},
  year={2020}
}