SMGNet

Official Pytorch Code base for "SMGNet:A Semantic Map-Guided Multitask Neural Network for Remote Sensing Image Semantic Change Detection"

Project

Introduction

We propose a novel multitask network (i.e., SMGNet) that fully integrates available semantic information derived from historical maps, with the aim of improving the under-detection of changed areas and misclassiffcation of changed classes derived from high-resolution satellite images.

Using the code:

The code is stable while using Python 3.9.0, CUDA >=11.0

• Clone this repository:

git clone https://github.com/long123524/SMGNet
cd SMGNet

To install all the dependencies using conda or pip:

PyTorch
TensorboardX
OpenCV
numpy
tqdm
skimage
...

Data Format

Make sure to put the files as the following structure:

inputs
└── <train>
    ├── image1
    |   ├── 001.tif
    │   ├── 002.tif
    │   ├── 003.tif
    │   ├── ...
    |
    └── image2
    |   ├── 001.tif
    |   ├── 002.tif
    |   ├── 003.tif
    |   ├── ...
    └── label1
    |   ├── 001.tif
    |   ├── 002.tif
    |   ├── 003.tif
    |   ├── ...
    └── label2
    |   ├── 001.tif
    |   ├── 002.tif
    |   ├── 003.tif
    |   ├── ...
    └── prior_t1
    |   ├── 001.tif
    |   ├── 002.tif
    |   ├── 003.tif
    |   ├── ...
    └── ...
    

For testing and validation datasets, the same structure as the above.

Datasets

A preprocessed data of HRSCD is available at https://pan.baidu.com/s/14FQNhSr-D8i1N3zeCqg0nQ code: m9u7

Training and testing

1. Train the model.

python train_SPG.py.

2. Test the model.

python pred_SCD.py.

3. Accuracy evaluation.

python Eval_SCD.py.

A pretrained weight

A pretrained weight of PVT-V2 on the ImageNet dataset is provided: https://drive.google.com/file/d/1uzeVfA4gEQ772vzLntnkqvWePSw84F6y/view?usp=sharing

Acknowledgements:

This code-base uses certain code-blocks and helper functions from HGINet and BiSRNet.

Citation:

If you find this work useful or interesting, please consider citing the following references.

@article{long2025,
  title={SMGNet:A Semantic Map-Guided Multitask Neural Network for Remote Sensing Image Semantic Change Detection},
  author={Long, Jiang and Liu, Sicong and Li, Mengmeng},
  journal={IEEE GEOSCIENCE AND REMOTE SENSING LETTERS},
  volume={22},
  pages={1--5},
  year={2025},
  publisher={IEEE}
}

@article{long2025,
  title={BGSNet: A boundary-guided Siamese multitask network for semantic change detection from high-resolution remote sensing images},
  author={Long, Jiang and Liu, Sicong and Li, Mengmeng and Zhao, Hang and Jin, Yanmin},
  journal={ISPRS Journal of Photogrammetry and Remote Sensing},
  volume={225},
  pages={221--237},
  year={2025},
  publisher={Elsevier}
}

@article{long2024,
  title={Semantic change detection using a hierarchical semantic graph interaction network from high-resolution remote sensing images},
  author={Long, Jiang and Li, Mengmeng and Wang, Xiaoqin and Stein, Alfred},
  journal={ISPRS Journal of Photogrammetry and Remote Sensing},
  volume={211},
  pages={318--335},
  year={2024},
  publisher={Elsevier}
}