Ocean Shield: Transformer-Based Detection of Marine Oil Spills from SAR Images

Marine oil spills pose serious environmental and economic threats, making fast and accurate detection essential. This research presents an intelligent oil spill detection framework based on a fine-tuned DINOv2 Vision Transformer applied to SAR satellite images. The model leverages self-supervised learning to extract robust features and captures both large-scale patterns and fine surface textures, enabling reliable discrimination between oil spills and similar natural phenomena such as low-wind areas, algal blooms, and natural slicks. Fine-tuning with labeled SAR datasets improves robustness across diverse sea conditions, weather scenarios, and satellite sensors, while attention-based visualizations enhance model interpretability. Experimental results demonstrate high detection accuracy, reduced false alarms, and strong generalization, making the proposed framework suitable for real-time marine surveillance, environmental monitoring, and disaster response applications.