wangqf / scdunetpp

Implementation for SCDUNet++

SCDUNet++ for Landslide Mapping Using Multi-Channel Remote Sensing Data

📄 This repository contains the official implementation of our paper: Landslide Mapping Based on a Hybrid CNN-Transformer Network and Deep Transfer Learning Using Remote Sensing Images with Topographic and Spectral Features Implementation for SCDUNet++.

📖 View paper here.

👉 Here, we provide the pytorch implementation of SCDUNet++.

Requirements

🖥️ This repo has been tested on Windows 10, Python 3.8, PyTorch 1.13.0, and CUDA 11.6. To setup the required modules, please run:

pip install -r requirements.txt

Installation

💾 Clone this repository to your local machine.

git clone https://github.com/lewuu/SCDUNetPP
cd SCDUNetPP

Dataset preparation

1️⃣ Create the directory 📁 data/luding/DATA/img and put the images in it. 🖼️
2️⃣ Create the directory 📁 data/luding/DATA/label and put the labels in it. 🏷️
3️⃣ Create the directory 📁 data/luding/DATA/config and put train.txt, val.txt and test.txt in it. 📝 The text files contain indexes of images and labels as follows (if the images are stored in TIF format):

img1.tif,label1.tif
img2.tif,label2.tif
...,...

🔧 In the file configs/config_case_luding.py, you have the flexibility to adjust the parameters based on your dataset requirements. For example, you can modify the num_classes, input_shape, and in_channels to suit your needs.

Train

✨ To initiate training, simply run the train.py. The weights for each epoch will be saved in the checkpoints/luding/scdunetpp/loss_yyyy_mm_dd_hh_mm_ss directory.

Test

✨ Select a weight for testing from the model weights saved during each epoch of training. Put it in checkpoints/Potsdam/scdunetpp directory, and change ckpt_test in configs/config_case_luding.py to the name of the selected weights file.

✨ Finally, run test.py to start testing.

Citation

If you find this repo useful for your research, please consider citing the paper as follows:

@article{WU2024103612,
title = {Landslide mapping based on a hybrid CNN-transformer network and deep transfer learning using remote sensing images with topographic and spectral features},
journal = {International Journal of Applied Earth Observation and Geoinformation},
volume = {126},
pages = {103612},
year = {2024},
issn = {1569-8432},
doi = {https://doi.org/10.1016/j.jag.2023.103612},
url = {https://www.sciencedirect.com/science/article/pii/S1569843223004363},
author = {Lei Wu and Rui Liu and Nengpan Ju and Ao Zhang and Jingsong Gou and Guolei He and Yuzhu Lei}
}