A Self-supervised-driven Open-set Unsupervised Domain Adaptation Method for Optical Remote Sensing Image Scene Classification and Retrieval

This repository is the official implementation of A Self-supervised-driven Open-set Unsupervised Domain Adaptation Method for Optical Remote Sensing Image Scene Classification and Retrieval (IEEE TGRS 2023).

<b>Authors</b>: Siyuan Wang, Dongyang Hou and Huaqiao Xing

Requirements

This code is written for python3.
pytorch >= 1.7.0
torchvision
numpy, prettytable, tqdm, scikit-learn, matplotlib, argparse, h5py

Data Preparing

Download dataset from the following link (code is xos6):

BaiduYun

Training and Evaluating

The pipeline for training with SSOUDA is the following (The code is still being optimized):

Train the model. For example, to run an experiment for UCM_LandUse dataset (source domain) and NWPU-RESISC45 dataset (target domain), run:

python ssouda.py /your_path/SSOUDA_dataset/ -s UCMD -t NWPU -a resnet50 --epochs 60 --seed 1 --log logs/ucmd_nwpu

Evaluate the model.

python ssouda.py /your_path/SSOUDA_dataset/ -s UCMD -t NWPU -a resnet50 --epochs 60 --seed 1 --log logs/ucmd_nwpu --phase test

Acknowledgment

This code is heavily borrowed from Transfer-Learning-Library

Citation

If you find our work useful in your research, please consider citing our paper:

@ARTICLE{10078892,
  author={Wang, Siyuan and Hou, Dongyang and Xing, Huaqiao},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={A Self-supervised-driven Open-set Unsupervised Domain Adaptation Method for Optical Remote Sensing Image Scene Classification and Retrieval}, 
  year={2023},
  doi={10.1109/TGRS.2023.3260873}
}

Contact

Please contact wsy.mail@foxmail.com if you have any question on the codes.

SSOUDA

Install / Use

README