Pixel-wise Agricultural Image Time Series Classification: Comparisons and a Deformable Prototype-based Approach

Elliot Vincent  Jean Ponce  Mathieu Aubry

Official PyTorch implementation of Pixel-wise Agricultural Image Time Series Classification: Comparisons and a Deformable Prototype-based Approach. Check out our webpage for other details!

If you find this code useful, don't forget to star the repo ⭐.

Installation ⚙️

1. Clone the repository in recursive mode

git clone [email protected]:ElliotVincent/AgriITSC.git --recursive

2. Create and activate virtual environment

python3 -m venv agriitsc
source agriitsc/bin/activate
python3 -m pip install -r requirements.txt

This implementation uses Pytorch.

How to use 🚀

We present steps to run our method on TimeSen2Crop, making our pre-pocessed version of this dataset available here:

• 🇦🇹 TimeSen2Crop [2]

To train and evaluate on other datasets, please follow the links below. All information on how we process the data is described in our paper.

• 🇫🇷 PASTIS [1]
• 🇿🇦 SA [4]
• 🇩🇪 DENETHOR [3]

1. Downlaoad the dataset

cd AgriITSC
mkdir datasets && cd datasets
gdown --id 1rCIyB4LETzfBhfYoc7dLHNKYhv8vJ315
unzip TimeSen2Crop.zip

2. Training the model

To train and evaluate our method with supervision do:

PYTHONPATH=$PYTHONPATH:./src python3 src/trainer.py -t supervised -c ts2c_dtits_supervised.yaml

And without supervision do:

PYTHONPATH=$PYTHONPATH:./src python3 src/trainer.py -t unsupervised -c ts2c_dtits_unsupervised.yaml

3. Saved model

Our trained models on TimeSen2Crop are available in results/, both for the supervised and unsupervised case.

Bibliography

[1] Vivien Sainte Fare Garnot et al. Panoptic segmentation of satellite image time series with convolutional temporal attention networks. ICCV, 2021.

[2] Giulio Weikmann et al. Timesen2crop: A million labeled samples dataset of sentinel 2 image time series for crop-type classification. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021.

[3] Lukas Kondmann et al. Denethor: The dynamicearthnet dataset for harmonized, inter-operable, analysis-ready, daily crop monitoring from space. NeurIPS Datasets and Benchmarks Track, 2021.

[4] Lukas Kondmann et al. Early crop type classification with satellite imagery: an empirical analysis. ICLR 3rd Workshop on Practical Machine Learning in Developing Countries, 2022.