A source-free unsupervised domain adaptation method for cross-regional and cross-time crop mapping from satellite image time series
The official implementation code for our paper "A source-free unsupervised domain adaptation method for cross-regional and cross-time crop mapping from satellite image time series".
Our paper has been accepted to Remote Sensing of Environment and is publicly available at: ScienceDirect.
1- Download the preprocessed data for the three sites in the USA from Google Drive - Then extract the ZIP file into the root directory.
2- Pretrain the model on the source domain data using the following code:
python source_training.py --pretrained_save_dir Pretrained_USA --backbone_network CNN --source_site A --source_year 2019 --data_dir Data_USA
-
backbone_network: You can select either 'CNN', 'Transformer', or 'LSTM'.
-
source_site: You can select either 'A','B',or 'C'. Please note that Sites A, B, and C correspond to Sites IA, MO, and MS in the paper, respectively.
-
source_year: You can select either '2019','2020',or '2021'.
-
pretrained_save_dir: Set the path for saving the pretrained model.
-
data_dir: Path where the data of the three sites are located.
3- Apply domain adaptation to the target domain data and evaluate the performance using the following code:
python AdaptationandEvaluation.py --adapted_save_dir Adapted --pretrained_save_dir Pretrained_USA --backbone_network CNN --source_site A --target_site C --source_year 2019 --target_year 2021
-
target_site: You can select either 'A','B',or 'C'.
-
target_year: You can select either '2019','2020',or '2021'.
-
adapted_save_dir: Set the path for saving the adapted models.
@article{mohammadi2024source,
title={A source-free unsupervised domain adaptation method for cross-regional and cross-time crop mapping from satellite image time series},
author={Mohammadi, Sina and Belgiu, Mariana and Stein, Alfred},
journal={Remote Sensing of Environment},
volume={314},
pages={114385},
year={2024},
publisher={Elsevier}
}