SLCNet

Pytorch code for the paper "Long-Range Correlation Supervision for Land-Cover Classification from Remote Sensing Images".

Environment

1. Python 3.7
2. pytorch>=1.5 torchvision>=0.6.0
3. opencv-python

Training

1.prepare your data

The organization of your data is as follows:
assuming your data is in the 'dir_path',
'dir_path' has a subfolder, i.e.'train_data'
'train_data' has two subfolders, i.e.'train' and 'val'
'train'and 'val' both have two subfolders, i.e., 'image' and 'label'

2.create a python file and input:

from train_model import train_multiclass_model

project_class_num=6
data_path='dir_path'
save_path='path for saving the training model weight'
batchsize=4
epoch=60
patch_size=512 #the size of your image/label tiles  
train_multiclass_model(project_class_num,data_path,save_path, batchsize, epoch,patch_size)

#run the python file and the SLCNet model start training

Predicting

1.prepare your data, just the images would be feded into the model

2.create a python file and input:

from predict_model import predict_multiclass_model

project_class_num=6
imgpath='test images path'#the sizes of the testing images can be random.
save_path='path for saving the predicting results'
weight_path='model weight path, .../*.pkl'
patch_size=512 #usually 512
predict_multiclass_model(project_class_num, imgpath,  weight_path,save_path, patch_size)

#run the python file and the SLCNet model start predicting

Citation link

If you find this project useful for your research, please cite this work.

Yu, D., & Ji, S. "Long-Range Correlation Supervision for Land-Cover Classification from Remote Sensing Images",IEEE Transactions on Geoscience and Remote Sensing (TGRS), vol. 61, pp. 1-14, 2023.

Update

1. 2023.10.13, We released the code of SCLNet.
2. 2023.10.15, We added the code for training and testing the SCLNet.

Contact: [email protected]. Any questions or discussions are welcomed!