MDAF-Net is a novel multimodal fusion framework designed for joint classification of hyperspectral imaging (HSI) and LiDAR data.
The proposed network integrates:
- Multi-scale feature extraction
- Adaptive spatial-channel interaction
- Frequency-aware fusion
to fully exploit complementary information across:
- Spatial domain
- Spectral domain
- Frequency domain
Extensive experiments demonstrate that MDAF-Net achieves state-of-the-art performance on multiple public remote sensing datasets.
We conducted 10 distinct data partitions based on IF_CALC implementation and adopted the average results across these iterations as the final reported outcomes in our study.
| Dataset | OA (%) | AA (%) | Kappa (%) |
|---|---|---|---|
| Houston | 96.02 | 96.63 | 95.70 |
| MUUFL | 85.61 | 85.08 | 81.39 |
| Trento | 99.51 | 98.96 | 99.34 |
To get started, we recommend setting up a conda environment and installing dependencies via pip. Use the following commands to set up your environment.
conda create -n mdafnet python==3.11
conda activate mdafnet
pip install -r requirements.txt
python demo.py
If this code is useful for your research, please cite this paper.
@ARTICLE{song2026multi,
title = {Multi-domain adaptive fusion network for multi-source remote sensing data classification},
author = {Song, Qiya and Peng, Jianle and Song, Weiwei and Sun, Bin and Dian, Renwei and Li, Shutao},
journal = {SCIENCE CHINA Information Sciences},
year = {2026},
}
We are deeply grateful to repositories IF_CALC, GLT and FDNet, which served as the foundational basis for our code implementation.