Cloud-Aware SAR Fusion for Enhanced Optical Sensing in Space Missions

This repository contains the official implementation for the paper "Cloud-Aware SAR Fusion for Enhanced Optical Sensing in Space Missions".

Overview

This project introduces a Cloud-Aware Reconstruction Framework for satellite image cloud removal. It fuses SAR and optical data, utilizing an attention mechanism and an adaptive loss strategy to prioritize and enhance reconstruction accuracy specifically in cloud-occluded regions, yielding high-fidelity, cloud-free optical images.

Prerequisites & Installation

This code was developed and tested with Python 3.10 and CUDA 12.4.

1. Create a Virtual Environment (Recommended):

   python -m venv cloud_removal_env
   .\cloud_removal_env\Scripts\activate # On Windows
   source cloud_removal_env/bin/activate # On Linux/macOS
   # Or, if using conda:
   # conda create -n cloud_removal_env python=3.10
   # conda activate cloud_removal_env

2. Install PyTorch with CUDA support: Ensure you install the PyTorch version compatible with your CUDA toolkit. For CUDA 12.4, use the following command (adjust if your CUDA version differs):

   pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124

3. Install Other Dependencies: After activating your virtual environment, navigate to the project's root directory (e.g., Cloud-Removal) and install the remaining libraries from requirements.txt:

   pip install -r requirements.txt

   (Note: Your requirements.txt file has been curated to include only direct dependencies, and pip will automatically handle their sub-dependencies.)

Dataset

This project utilizes the SEN12MS-CR DATASET, which serves as a benchmark for multimodal cloud removal in remote sensing.

Get Started

1. Train the Network

cd codes/
python main_train.py

To view more training options:

cd codes/
python main_train.py --help

2. Test the Network

cd codes/
python main_test.py --model_path ../ckpt/CR_net.pth

Test results and metric summaries will be saved in the ./results/ directory.

Contact

Email: thoailt@hcmue.edu.vn