Skip to content
Temporal Convolutional Neural Network for the Classification of Satellite Image Time Series
Branch: master
Clone or download
Latest commit f92feff Jul 16, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
deeplearning NEW: first code version Nov 26, 2018
map NEW: maps Jul 16, 2019
outputfiles NEW: first code version Nov 26, 2018
sits NEW: first code version Nov 26, 2018
LICENSE Initial commit Nov 21, 2018
README.md UPD Jul 16, 2019
run_archi.py WIP: writting README.md file Nov 27, 2018
run_main_archi.py WIP: writting README.md file Nov 27, 2018
test_dataset.csv NEW: first code version Nov 26, 2018
train_dataset.csv NEW: first code version Nov 26, 2018

README.md

Temporal Convolutional Neural Network

Training temporal Convolution Neural Netoworks (CNNs) on satelitte image time series.
This code is supporting by a paper published in Remote Sensing:

@article{Pelletier2019Temporal,
  title={Temporal convolutional neural network for the classification of satellite image time series},
  author={Pelletier, Charlotte and Webb, Geoffrey I and Petitjean, Fran{\c{c}}ois},
  journal={Remote Sensing},
  volume={11},
  number={5},
  pages={523},
  year={2019},
  publisher={Multidisciplinary Digital Publishing Institute},
  note={https://www.mdpi.com/2072-4292/11/5/523}
}

Prerequisites

This code relies on Pyhton 3.6 (and should work on Python 2.7) and Keras with Tensorflow backend.

Examples

Running the models

  • main architecture: python run_main_archi.py
  • other experiments described in the related paper:
python run_archi.py --sits_path ./ --res_path path/to/results --noarchi 0

The architecture will run by training the network on train_dataset.csv file and by testing it on test_dataset.csv file.
Please note that both train_dataset.csv and test_dataset.csv files are a subsample of the data used in the paper: original data cannot be distributed.

Thoses files have no header, and contain one observation per row having the following format: [class,date1.NIR,date1.R,date1.G,date2.NIR,...,date149.G]

Changing network parameters

  • Number of channels in the data: n_channels = 3 (run_archi.py, L21).
    It will require to change functions contained in readingsits.py.
  • Validation rate: val_rate = 0.05 (run_archi.py, L22).
  • Network hyperparameters are mainly defined in architecture_features.py file.

Maps

The produced map for TempCNNs and RFs are available in the map folder.

Contributors

You can’t perform that action at this time.