MCD-Mamba

MCD-Mamba: Mamba-Based Framework for Infrastructure Change Detection using Multimodal Satellite Imagery

Mira Kenzhebay *, Reza Arghandeh

Western Norway University of Applied Sciences

^* Corresponding author

🔭Overview

Change detection using remote sensing data has shown remarkable progress through deep learning approaches, particularly with Convolutional Neural Networks (CNNs) and Transformers, yet both architectures face limitations either in terms of computational efficiency or accuracy. This paper introduces MCD-Mamba (Multimodal Change Detection Mamba), a novel framework that leverages state space modeling for efficient multimodal change detection using both optical and Synthetic Aperture Radar (SAR) data. The architecture features modality-specific Visual Mamba (VMamba) encoders with strategic weight sharing, an efficient multi-scale fusion strategy, and a novel Difference-based Spatio-Temporal (DST) block for modeling temporal changes. Experiments conducted on the Multimodal ONERA Satellite Change Detection (OSCD) dataset demonstrate that MCD-Mamba outperforms existing architectures, achieving a 61.01% F-score, 65.96% recall, and 81.30% balanced accuracy. The model strikes an optimal balance between performance and efficiency, requiring moderate computational resources (7.1M parameters) while maintaining faster convergence compared to CNN and Transformer alternatives. Notably, a compact variant of the architecture achieves comparable performance while using only 4% of the base model's parameters. Visual analysis reveals that MCD-Mamba exhibits superior sensitivity in urban environments and better captures fine-grained changes in scattered urban structures compared to existing methods. These results demonstrate the potential of state space modeling for multi-modal remote sensing applications, offering a promising direction for efficient and accurate infrastructure change detection.

🗝️Let's Get Started!

A. Installation

The repo is based on the VMama repo and MambaCD. The following installation sequence is taken from the VMamba repo.

Step 1: Clone the repository:

Clone this repository and navigate to the project directory:

git clone https://github.com/ci2lab/MCD.git
cd MambaCD

Step 2: Environment Setup:

It is recommended to set up a conda environment and installing dependencies via pip. Use the following commands to set up your environment:

Create and activate a new conda environment

conda create -n changemamba
conda activate changemamba

Install dependencies

pip install -r requirements_upd.txt
cd kernels/selective_scan && pip install .

Dependencies for "Detection" and "Segmentation" (optional in VMamba)

pip install mmengine==0.10.1 mmcv==2.1.0 opencv-python-headless ftfy regex
pip install mmdet==3.3.0 mmsegmentation==1.2.2 mmpretrain==1.2.0

B. Data Preparation

Binary change detection The OSCD is used for binary change detection experiments. Please download it and make it have the following folder/file structure:

${DATASET_ROOT}   # Dataset root directory, for example: /home/mira/Data/Onera
├───S1
│   ├───abudhabi
│   │   ├───imgs_1
│   │   │   ├───preview
│   │   │   └───transformed
│   │   │           S1A_IW_GRDH_1SDV_20160218T142406_20160218T142442_010002_00EB50_77FE.tif
│   │   └───imgs_2
│   │       ├───preview
│   │       └───transformed
│   │               S1A_IW_GRDH_1SDV_20180327T142413_20180327T142438_021202_024756_B24C.tif
│   ├───aguasclaras
│   │   ├───....
│   ├───....
├───S2
│   ├───abudhabi
│   │   ├───imgs_1
│   │   ├───imgs_1_rect
│   │   │       B01.tif
│   │   │       B02.tif
│   │   │       B03.tif
│   │   │       ....
│   │   ├───imgs_2
│   │   ├───imgs_2_rect
│   │   │       B01.tif
│   │   │       B02.tif
│   │   │       B03.tif
│   │   │       ....
│   │   └───pair
│   ├───aguasclaras
│   │   ├───...
│   ├───....
├───Train Labels
│   ├───abudhabi
│   │   └───cm
│   │           abudhabi-cm.tif
│   │           cm.png
│   ├───aguasclaras
│   │   └───...
    ├───....
└───Test Labels
    ├───brasilia
    │   └───cm
    │           brasilia-cm.tif
    │           cm.png
    ├───chongqing
    │   └───...
    ├───....

C. Model Training

To train models, change working directory to [changedetection] folder, which contains all the code for network definitions, training and testing.

cd <project_path>/MCD-Mamba/changedetection

Binary change detection

The following commands are examples of how to train and evaluate MCD-Mamba-compact on the multimodal OSCD dataset:

python3 changedetection/script/train_ONERA.py --cfg 'changedetection/configs/vssm1/vssm_compact_224.yaml' 
                                              --dataset_path /content/drive/MyDrive/Data/ONERA   
                                              --model_param_path 'changedetection/saved_models' 
                                              --patch_size 224  
                                              --type train 
                                              --model 'STMambaBCD_multimodal' 
                                              --dataset 'ONERA-multimodal' 
                                              --epochs 100 
                                              --learning_rate 1e-3 
                                              --batch_size 16 
                                              --lovasz_weight 0.7 
                                              --opt_bands 13


### `D. Inference Using Saved Model'

python3 changedetection/script/infer_ONERA.py --cfg 'changedetection/configs/vssm1/vssm_compact_224.yaml' 
                                              --dataset_path /content/drive/MyDrive/Data/ONERA   
                                              --patch_size 224  
                                              --type test 
                                              --model 'STMambaBCD_multimodal' 
                                              --dataset 'ONERA-multimodal' 
                                              --resume 'changedetection/saved_models/ONERA-multimodal/STMambaBCD_multimodal/best_model.pth' 
                                              --result_saved_path 'changedetection/saved_models/ONERA-multimodal/STMambaBCD_multimodal'

📜Reference

@article{
}

🤝Acknowledgments

This project is based on MambaCD, VMamba (paper, code), ScanNet (paper, code), BDANet (paper, code). Big thanks for their brilliant works!!

🙋Q & A

If you have any questions, please feel free to contact us.