Home

Awesome

s1_icetype_cnn-1.1

Retrieve sea ice types from corrected Sentinel-1 SAR with convolutional neural networks.

Run the example Jupyter Notebook

  1. Start the official Tensorflow Docker container:
docker run -v $PWD:/tf --rm -it -p 8888:8888  tensorflow/tensorflow:latest-jupyter
  1. Open the provided URL in a browser and open the notebook "cnn_for_sea_ice_types.ipynb"
  2. Execute the cells

Installation and usage on any platform using Docker

  1. Build image s1denoised as explained here: https://github.com/nansencenter/sentinel1denoised/

  2. Build image s1icetype using Dockerfile from this repository: docker build . -t s1icetype

  3. Run thermal noise, texture noise and angular corrections

docker run --rm -it \
    -v /path/to/s1/files:/data \
    s1icetype \
    s1_correction.py \
    /data/S1B_EW_GRDM_1SDH_20200323T053203_20200323T053303_020815_027780_8729.zip \
    /data/S1B_EW_GRDM_1SDH_20200323T053203.tif
  1. Run the ice type algorithm (mounted directory names can be anything)
docker run --rm -it \
    -v /path/to/cnn/file:/cnn \
    -v /path/to/s1/files:/data \
    s1icetype \
    get_icetype.py /cnn/cnn_weights.hdf5 \
    /data/S1B_EW_GRDM_1SDH_20200323T053203.tif /data/S1B_EW_GRDM_1SDH_20200323T053203_icetype.tif

Installation and usage on Linux using conda

  1. Create environment s1denoise for S1 correction as explained here: https://github.com/nansencenter/sentinel1denoised/

  2. Activate environment s1denoise and install tensorflow 2.1.0

conda activate s1denoise
conda install tensorflow=2.1.0
  1. Run thermal noise, texture noise and angular corrections
s1_correction.py S1B_EW_GRDM_1SDH_20200323T053203_20200323T053303_020815_027780_8729.zip S1B_EW_GRDM_1SDH_20200323T053203.tif
  1. Run ice type algorithm
./get_icetype.py cnn_weights.hdf5 S1B_EW_GRDM_1SDH_20200323T053203.tif S1B_EW_GRDM_1SDH_20200323T053203_icetype.tif