Awesome

<h1 align="center">Unsupervised Change Detection in Multitemporal VHR Images Based on Deep Kernel PCA Convolutional Mapping Network</h1> <h3 align="center"> <a href="https://scholar.google.com/citations?user=DbTt_CcAAAAJ&hl=zh-CN">Chen Wu</a>, <a href="https://chrx97.com/">Hongruixuan Chen</a>, <a href="https://scholar.google.com/citations?user=Shy1gnMAAAAJ&hl=zh-CN">Bo Du</a>, and <a href="https://scholar.google.com/citations?user=vzj2hcYAAAAJ&hl=en">Liangpei Zhang</a></h3>

This is an official implementation of unsupervised multi-class change detection framework KPCAMNet in our IEEE TCYB 2022 paper: Unsupervised Change Detection in Multitemporal VHR Images Based on Deep Kernel PCA Convolutional Mapping Network.

Abstract

With the development of Earth observation technology, very-high-resolution (VHR) image has become an important data source of change detection. Nowadays, deep learning methods have achieved conspicuous performance in the change detection of VHR images. Nonetheless, most of the existing change detection models based on deep learning require annotated training samples. In this paper, a novel unsupervised model called kernel principal component analysis (KPCA) convolution is proposed for extracting representative features from multi-temporal VHR images. Based on the KPCA convolution, an unsupervised deep siamese KPCA convolutional mapping network (KPCA-MNet) is designed for binary and multi-class change detection. In the KPCA-MNet, the high-level spatial-spectral feature maps are extracted by a deep siamese network consisting of weight-shared KPCA convolutional layers. Then, the change information in the feature difference map is mapped into a 2-D polar domain. Finally, the change detection results are generated by threshold segmentation and clustering algorithms. All procedures of KPCA-MNet do not require labeled data. The theoretical analysis and experimental results in two binary change detection datasets and one multi-class change detection datasets demonstrate the validity, robustness, and potential of the proposed method.

Network architecture

Get started

Requirements

python==3.6.12
scikit-learn==0.19.0
imageio=2.9.0
numpy==1.13.3
gdal==2.2.2

Datasets

Two binary change detection datasets, WH and QU, and one multi-class change detection dataset, HY, are used in our work. You can download them here for reproducibility or your own research. [<a href="https://drive.google.com/file/d/1kR4yNafYdhuGSo4kPI1HwApmluY3L1Oq/view?usp=drive_link">GoogleDrive</a>], [<a href="https://pan.baidu.com/s/1WF3wTtZhZ-5hUul4UrLETg?pwd=jg6s">BaiduCloud</a>]

Dataset	Pre-event image	Post-event image	Reference Image
WH	<img src="./figures/WH_1.png" width=250px>	<img src="./figures/WH_1.png" width=250px>	<img src="./figures/WH_GT.png" width=250px>
QU	<img src="./figures/QU_1.jpg" width=250px>	<img src="./figures/QU_2.jpg" width=250px>	<img src="./figures/QU_GT.jpg" width=250>
HY	<img src="./figures/HY_1.png" width=250px>	<img src="./figures/HY_2.png" width=250px>	<img src="./figures/change_ref.png" width=250px>

</center>

Train & infer

python train_KPCAMNet.py

Detection results

Predictions on the Hanyang dataset:

Pre-event image	Post-event image	Reference	Prediction
<img src="./figures/HY_1.png" width=250px>	<img src="./figures/HY_2.png" width=250px>	<img src="./figures/change_ref.png" width=250px >	<img src="./figures/change_map.png" width=250px >

</center>

The results of quantitative assessments:

Citation

If this code or dataset contributes to your research, please kindly consider citing our paper. We appreciate your support!🙂

@article{Wu2022Unsupervised,
  author={Wu, Chen and Chen, Hongruixuan and Du, Bo and Zhang, Liangpei},
  journal={IEEE Transactions on Cybernetics}, 
  title={Unsupervised Change Detection in Multitemporal VHR Images Based on Deep Kernel PCA Convolutional Mapping Network}, 
  year={2022},
  volume={52},
  number={11},
  pages={12084-12098},
  doi={10.1109/TCYB.2021.3086884}
}

Q & A

For any questions, please contact us.