WTS: A weakly towards strongly supervised learning framework for remote sensing land cover classification using segmentation models
This is the implement of WTS supervised learning framework for remote sensing land cover classification using segmentation models, in which the SRG algorithm is referring to https://github.com/xtudbxk/DSRG-tensorflow.
If you find this code is useful for your research, please consider citing it:
@article{wts,
title={WTS: A weakly towards strongly supervised learning framework for remote sensing land cover classification using segmentation models},
author={Wei Zhang, Ping Tang, Thomas Corpetti and Lijun Zhao},
booktitle={Remote Sensing},
pages={},
year={2021}
}
We tested the code on
- keras 2.2
- tensorflow 1.11
- python 3.6
Other dependencies:
- numpy
- tqdm
- gdal
- cv2
- segmentation_models
- matplotlib
- sklearn
- pydensecrf
-Train SVM and generate initial seed
python generate_initial_seed.py
-Train Segmentation model and update seed iteratively
python train_update_seed.py