This repository contains an example of how to train the deep learning architecture and how to use the interpretability tools used in the paper Recognition of polar lows in Sentinel-1 SAR images with deep learning by J. Grahn and F. M. Bianchi.

Dataset

The Sentinel-1 maritime mesocyclone dataset is publicly available and can be downloaded here.

Installation (with Anaconda)

Create an Anaconda environment using the environment.yml file.

conda env create -f environment.yml

The environment was created on Ubuntu 20.04. For more details on how to create and manage an environment, look here.

Pre-trained models

The best-performing models can be downloaded here.

To use a pre-trained model:

import tensorflow as tf
import tensorflow_addons as tfa

model = tf.keras.models.load_model(
    "models/model_1536_F1_095.h5", 
    custom_objects={'AdamW': tfa.optimizers.AdamW})
img = tf.keras.preprocessing.image.load_img(
    "data/test/pos/cffe42_20191012T084028_20191012T084212_mos_rgb.png", 
    target_size=(1536, 1536))
img_array = tf.keras.preprocessing.image.img_to_array(img)

pred = model.predict(tf.expand_dims(img_array, 0))
print("Predicted class: ", pred[0])

Model training from scratch

If you want to train the deep learning model from scratch take a look at train_model.py, which provides a simple example of how to train the architecture adopted in our paper. The script downloads automatically the dataset in the data/ folder.

Model interpretability

The following notebooks show how to use the interpretability techniques to see what the deep learning model is focusing on.

• GradCAM.ipynb
• Integrated_gradients.ipynb

Citation

Please, consider citing our paper if you are using our dataset in your research.

@article{grahn2022recognition,
  title={Recognition of polar lows in Sentinel-1 SAR images with deep learning},
  author={Grahn, Jakob and Bianchi, Filippo Maria},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  volume={60},
  pages={1--12},
  year={2022},
  publisher={IEEE}
}