This repository presents a ship detection model built upon the YOLOv7 architecture. To enhance the performance of the model, we've incorporated a preprocessing pipeline involving DDPM and anisotropic diffusion to mitigate noise prevalent in SAR images. The model is trained on a combination of SSDD, DS SDD, and Fusar datasets.
Key Components
- Noise Reduction: DDPM and anisotropic diffusion are employed to preprocess SAR images, significantly reducing noise and improving image quality.
- Object Detection: YOLOv7 serves as the backbone for detecting ships within the preprocessed images, providing accurate bounding boxes and confidence scores.
- Dataset: The model is trained on a merged dataset comprised of SSDD, DS SDD, and Fusar, ensuring robustness and diversity in ship appearance.
Future Work
- Explore different noise reduction techniques (e.g., wavelet denoising)
- Incorporate additional datasets for increased model diversity
- Optimize YOLOv7 architecture for SAR image processing
Contribution
We welcome contributions to this project. Please feel free to submit pull requests with improvements or bug fixes.