MCAMamba: Multi-scale Feature Fusion State Space Model for Multi-source Remote Sensing Image Classification

---

📌 Introduction

MCAMamba is a Multi-Level Cross-Modal Attention-Guided State Space Model for multi-source remote sensing image classification. It addresses three critical challenges: insufficient spatial-spectral coordination, cross-modal heterogeneity, and inadequate multi-scale feature integration. Through a novel three-stage feature fusion pipeline (FExt-Attention, SSM-Attention, and FFus-Attention), MCAMamba achieves superior feature representation, demonstrating state-of-the-art performance on HSI-LiDAR/SAR classification tasks.

📂 Dataset

The datasets used in our experiments can be obtained from the following link:
📥 Download Houston2013 Dataset

---

🛠 Installation and Dependencies

Before running the code, please ensure the following dependencies are installed:

pip install causal-conv1d==1.1.1
pip install mamba-ssm==1.0.1

---

🏋️‍♂️ Usage: Training MCAMamba

To train MCAMamba on the Houston2013 dataset, use the following command:

python train.py --epoch 40 --lr 1e-4 --batchsize 128 --dataset Houston2013

📬 Contact

For any questions, please contact us via email:
📧 doumingyu24@mails.ucas.ac.cn

---

📚 Citation

If you use MCAMamba in your research, please cite our work:

@ARTICLE{11194194,
  author={Dou, Mingyu and Qiu, Shi and Hu, Ming and Qiao, Xiaozhen and Ye, Huping and Liao, Xiaohan and Sun, Zhe},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={MCAMamba: Multi-Level Cross-Modal Attention-Guided State Space Model for Multi-Source Remote Sensing Image Classification}, 
  year={2025},
  volume={},
  number={},
  pages={1-1},
  keywords={Remote sensing;Feature extraction;Computational modeling;Transformers;Laser radar;Computational efficiency;Adaptation models;Synthetic aperture radar;Faces;Computational complexity;state space model;attention mechanism;multi-source remote sensing classification;hyperspectral image;synthetic aperture radar;LiDAR},
  doi={10.1109/TGRS.2025.3618301}}