Cross-supervised learning for cloud detection

The repository is for the article "cross-supervised learning for cloud detection" on GIScience&Remote Sensing.

The current version on the website does't include Supplementary. Download the Supplementary at here.

Requirements

• Python 3.6.13
• Pytorch 1.8.1
• Gdal
• VisualDL

Train

1. config the training details in the file: configs/MyNet_GF1.py
2. In terminal, train the nets by:

   python train.py
   or with specific GPUs (X for the index of GPUs):

   CUDA_VISIBLE_DEVICES=X,X,X python train.py

Test

1. config the model_name, dataset, pth1, pth2 and exp_id in the file: test.py.
2. In terminal, inference the nets by:

   python test.py
   or with specific GPUs (X for the index of GPUs):

   CUDA_VISIBLE_DEVICES=X,X,X python test.py

HY1C-UPC Dataset

HY1C-UPC Dataset is built from images of Chinese HY1-C satellite. The coastal zone imager (CZI) on the HY1-C satellite has a 50-m spatial resolution with four multi-spectral bands. The HY1C-UPC dataset includes 25 scenes from September 2021 to February 2022. The main scenes are collected from the coastal zones as shown bellow. The observation width of the CZI is large, hence, the dataset includes various terrains, e.g., city, snow, forest, ocean, etc., as shown bellow. The HY1C-UPC dataset contains 8 manually labeled scenes that are labeled by experts with the Photoshop software and 17 unlabeled scenes. HY1C-UPC dadaset is avaliable at: aliyundrive. (key: uu49)

Todo

• Test code release
• Train code release
• HY1C-UPC dataset release

Citation

If you use this project in your research please cite:

@article{CSL:WU2023,
    doi = {10.1080/15481603.2022.2147298},
    author = {Kang Wu and Zunxiao Xu and Xinrong Lyu and Peng Ren},
    title = {Cross-supervised learning for cloud detection},
    journal = {GIScience \& Remote Sensing},
    volume = {60},
    number = {1},
    pages = {2147298},
    year  = {2023},
    publisher = {Taylor & Francis},
    doi = {10.1080/15481603.2022.2147298},
    URL = {https://doi.org/10.1080/15481603.2022.2147298},
    eprint = {https://doi.org/10.1080/15481603.2022.2147298}
}

Ackonwledgement

Some implementations are built based on segmentation_models.pytorch.

Disclaimer

This repository can only be used for personal/research/non-commercial purposes. If you have any questions about this work, please raise an issue or contact me at kang_wu#foxmail.com. (Please replace # with @)