Robust feature aggregation network for lightweight and effective remote sensing image change detection (ISPRS 2024)
Authors: Zhi-Hui You; Si-Bao Chen; Jia-Xin Wang; Bin Luo
This repository contains simple pytorch implementation of our paper RFANet.
A lightweight change detection network, called as robust feature aggregation network (RFANet). To improve representative capability of weaker features extracted from lightweight backbone, a feature reinforcement module (FRM) is proposed. FRM allows current level feature to densely interact and fuse with other level features, thus accomplishing the complementarity of fine-grained details and semantic information. Considering massive objects with rich correlations in RS images, we design semantic split-aggregation module (SSAM) to better capture global semantic information of changed objects. Besides, we present a lightweight decoder containing channel interaction module (CIM), which allows multi-level refined difference features to emphasize changed areas and suppress background and pseudo-changes.
-
Prepare the data: Download the change detection datasets from the following links. Place them inside your
datasets
folder.LEVIR-CD
WHU-CD
CDD-CD
SYSU-CD
./samples/test
is a sample to start quickly.
- Crop all datasets into 256x256 patches.
- Generate list file as
ls -R ./label/* > test.txt
- Prepare datasets into following structure and set their path in
train.py
andtest.py
├─A ├─A1.jpg/png ├─A2.jpg/png ├─...jpg/png └─...jpg/png ├─B ├─B1.jpg/png ├─B2.jpg/png ├─...jpg/png └─...jpg/png ├─label ├─label1.jpg/png ├─label2.jpg/png ├─...jpg/png └─...jpg/png ├─list ├─train.txt ├─val.txt └─test.txt
- Prerequisites for Python:
- Creating a virtual environment in terminal:
conda create -n RFANet python=3.8
- Installing necessary packages:
pip install -r requirements.txt
- Creating a virtual environment in terminal:
- Clone this repo:
git clone https://github.com/Youzhihui/RFANet.git cd RFANet
- Train/Test
sh ./train_test_tools/train.sh
sh ./train_test_tools/test.sh
The quantitative results of different CD methods on LEVIR-CD, WHU-CD, and CDD-CD. The last column indicates the mean value of same report on different datasets. Color convention: best (red-bold), 2nd-best (blue-underline), and 3nd-best (orange-italic).
Heatmap visualizations of multiple levels for RFANet.
This repository is built under the help of the projects A2Net, BIT_CD, CDLab, and MobileSal for academic use only.
Please cite our paper if you find the work useful:
@article{you2024robust,
title={Robust feature aggregation network for lightweight and effective remote sensing image change detection},
author={You, Zhi-Hui and Chen, Si-Bao and Wang, Jia-Xin and Luo, Bin},
journal={ISPRS Journal of Photogrammetry and Remote Sensing},
volume={215},
pages={31--43},
year={2024},
publisher={Elsevier}
}