CHeGCN

Implementation for: CHeGCN (CNN-enhanced Hetergeneous Graph Convolutional Networks)

New work

Here is a new work of semantic segmention with GCN, published in International Journal of Digital Earth. The code is released at https://github.com/Liuzhizhiooo/GHGCN.

Software

package installation

pip install -r ./requirements.txt

Train & Evaluate

train models

# HoGCN
python main.py train --train=True --dataset=datasetHeGCN --model=HoGCN --tag=HoGCN-1

# HeGCN
python main.py train --train=True --dataset=datasetHeGCN --model=HeGCN --tag=HeGCN-1

# CHoGCN
python main.py train --train=True --dataset=datasetCHeGCN --model=CHeGCN --isGCNLabel=False --tag=CHoGCN-1

# CHeGCN
python main.py train --train=True --dataset=datasetCHeGCN --model=CHeGCN --isGCNLabel=True --tag=CHeGCN-1

test models

You may need to select the best model based on the validation OA curve and set the testModel, the epochs_90.pth is just an example.

# HoGCN
python main.py test --dataset=datasetHeGCN --model=HoGCN --tag=HoGCN-1 --testModel=epochs_90.pth

# HeGCN
python main.py test --dataset=datasetHeGCN --model=HeGCN --tag=HeGCN-1 --testModel=epochs_90.pth

# CHoGCN
python main.py test --dataset=datasetCHeGCN --model=CHeGCN --isGCNLabel=False --tag=CHoGCN-1 --testModel=epochs_90.pth

# CHeGCN
python main.py test --dataset=datasetCHeGCN --model=CHeGCN --isGCNLabel=True --tag=CHeGCN-1 --testModel=epochs_90.pth

Best Model

We provide the best model trained from the Beijing dataset in ./outputs/CHeGCN/checkpoints/epochs_70.pth

The visualization results of the 3-layer CHeGCN in our paper can be reproduced by the following command:

the segmentation parameters n=50,c=10.

python main.py test --dataset=datasetCHeGCN --model=CHeGCN --isGCNLabel=True --tag=CHeGCN --testModel=epochs_70.pth

Datasets

The Beijing dataset and Shenzhen dataset can be downloaded from Baidu drive or Google drive.

• Link: Baidu drive (extraction code: runc)
• Link: Google drive

The details about these datasets can be found in our article.

./dataset

├─divide  # data division
│  │  train.txt  # training
│  │  val.txt  # validation
│  │  test.txt  # test
│  │  
│  └─shp  # dataset distribution
│          
├─img  # images
│  ├─slide  # init images
│  │      
│  └─slideEnhanced  # images after space augmentation (slide window)
│
├─landcover  # land cover
│
├─park  # park label
│
├─npy  # graph data
│  ├─adjacency  # adjacency matrix
│  ├─node  # features of nodes
│  ├─landcover  # node-level land cover map
│  └─park  # node-level park label
│
└─SLICResult  # class-wise segmentation of images by SLIC
   └─segments

Class-wise SLIC Method

This class-wise SLIC method makes use of image texture to achieve clustering and guarantees that all pixels in each superpixel belong to the same land cover category.

Citation

If this code helps in your work, please consider citing us.

Liu, Z.-Q.; Tang, P.; Zhang, W.; Zhang, Z. CNN-Enhanced Heterogeneous Graph Convolutional Network: Inferring Land Use from Land Cover with a Case Study of Park Segmentation. Remote Sens. 2022, 14, 5027. https://doi.org/10.3390/rs14195027

or

@Article{rs14195027,
AUTHOR = {Liu, Zhi-Qiang and Tang, Ping and Zhang, Weixiong and Zhang, Zheng},
TITLE = {CNN-Enhanced Heterogeneous Graph Convolutional Network: Inferring Land Use from Land Cover with a Case Study of Park Segmentation},
JOURNAL = {Remote Sensing},
VOLUME = {14},
YEAR = {2022},
NUMBER = {19},
ARTICLE-NUMBER = {5027},
URL = {https://www.mdpi.com/2072-4292/14/19/5027},
ISSN = {2072-4292},
DOI = {10.3390/rs14195027}
}

License

The code and the models are MIT licensed, as found in the LICENSE file.