AAP: Adversarial Attenuation Patch against SAR Object Detection Model

This is the official PyTorch implementation of the AAP (Adversarial Attenuation Patch) method and integrates with popular attack algorithms from the Adversarial Robustness Toolbox (ART).

📌 Key Features

• AAP Method: A novel Adversarial Attenuation Patch approach for SAR object detection.
• Multiple Benchmark Methods: Supports standard attacks including AdvPatch, DPatch, and RobustDPatch via ART.
• YOLO Integration: Seamless integration with YOLO-based object detection models.
• Modular Design: Clean and extensible architecture for research and experimentation.
• Evaluation Tools: Built-in metrics and visualization for assessing attack performance.

📁 Project Structure

SAAP/
├── aap_utils/                    # Core utilities and custom implementations
│   ├── adv_attenuation_patch.py  # AAP (Adversarial Attenuation Patch) implementation
│   ├── aap_utils.py             # Helper functions and model wrappers
│   ├── random_attack.py         # Random attack utilities for toy experiments
│   └── random_pipeline.py       # Attack pipeline management
├── adv_examples/                # Directory for adversarial examples (empty by default)
├── art/                         # Minimal integrated components from the Adversarial Robustness Toolbox (ART)
│   ├── attacks/
│   │   └── evasion/
│   │       ├── adversarial_patch/    # Standard adversarial patch attack
│   │       ├── dpatch.py             # DPatch attack implementation
│   │       └── dpatch_robust.py      # Robust DPatch attack implementation
│   ├── estimators/              # Model estimators (PyTorch YOLO support)
│   └── ...                      # Other essential ART components
├── dataset/                     # Dataset directory (empty by default)
├── models/                      # YOLO model configurations
│   └── hub/                     # Predefined YOLO architectures
├── result_pipeline/             # Evaluation results directory (empty by default)
├── utils/                       # General utilities and tools
├── weights/                     # Model weights directory (empty by default)
├── attack_advpatch.py          # Standard adversarial patch attack script
├── attack_dpatch.py            # DPatch attack script
├── attack_robustdpatch.py      # RobustDPatch attack script
├── attackv1.py                 # Integrated development version (for learning ART interfaces)
├── attackv2.py                 # Standard production version (using AAP method)
├── detect_ship.py              # Ship detection example
├── val.py                      # Validation and evaluation script
└── advset.yaml                 # Dataset configuration

🚀 Getting Started

Prerequisites

• Python 3.8+
• PyTorch 2.2.2
• YOLOv5 (Ultralytics implementation)
• CUDA (optional, for GPU acceleration)

Installation

# Clone the repository
git clone https://github.com/your-username/SAAP.git
cd SAAP

# Install dependencies
pip install -r requirements.txt

Usage Examples

Run supported attacking methods

python attack_advpatch.py
python attack_dpatch.py
python attack_robustdpatch.py
python attackv2.py  # AAP method

📊 Evaluation

The platform provides comprehensive evaluation metrics including:

• False Negative Rate (FNR)
• False Positives Per Image (FPPI)
• mAP and detection confidence reduction

Run evaluation with:

python result_fprfnr.py
python result_pipeline.py

🤝 Contributing

Contributions are welcome! Please feel free to submit issues, feature requests, or pull requests.

📜 License

This project is licensed under the MIT License - see the LICENSE file for details.

📚 Citation

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

@article{Zhang2026SAAP,
  title   = {Towards Physically Realizable Adversarial Attenuation Patch against SAR Object Detection},
  author  = {Yiming Zhang and Weibo Qin and Feng Wang},
  journal = {arXiv preprint arXiv:2604.00887},
  year    = {2026}
}

🔒 Disclaimer

This code is intended for research and educational purposes only. Users are responsible for ensuring compliance with applicable laws and regulations when using adversarial attack techniques.

Acknowledgments

• This project makes extensive use of the Adversarial Robustness Toolbox (ART) for implementing standard adversarial patch methods, including AdvPatch, DPatch, and RobustDPatch.
• The object detection backbone is based on Ultralytics YOLOv5.