This is the source codebase for: Deep Evidential Remote Sensing Landslide Image Classification With a New Divergence, Multi-Scale Saliency and an Improved Three-Branched Fusion. This work has been accepted by IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (IEEE JSTARS) in Jan 12 2024 at https://doi.org/10.1109/JSTARS.2024.3354455. We sincerely thank the timely help of Prof. Lianmeng Jiao from Northwestern Polytechnical University (NPU) for improving the paper's quality.
Graphical abstract:
Post:
HOW TO USE THIS REPO: The experiments of this paper can be repeated on your platform with the following steps:
-
Download the Bijie landslide image dataset, which is avaliable at http://gpcv.whu.edu.cn/data/Bijie_pages.html *.
-
Use the code multiscale_saliency_grad.m to obtain multi-scale visual saliency. Then the remote sensing images after channel-wise fusion can be avaliable.
-
Train the deep neural networks using fused images with the source code (in the folder Awesome-Backbones-main, which is from repo https://github.com/Fafa-DL/Awesome-Backbones) and initial weights in OpenMMLab's Image Classification Toolbox and Benchmark, which is avaliable at https://github.com/open-mmlab/mmclassification.
-
Fuse the decisions from deep neural networks with the proposed three-branched fusion model (network_fusion.py).
Please cite this paper if this code contributes to your research:
@article{zhang2024deep,
title={Deep Evidential Remote Sensing Landslide Image Classification With a New Divergence, Multi-Scale Saliency and an Improved Three-Branched Fusion},
author={Zhang, Jiaxu and Cui, Qi and Ma, Xiaojian},
journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing},
volume={69},
pages={3799-3820},
year={2024}
}