Deep Learning on Sentinel-1 SAR Imagery for Ship Detection Using YOLO-V8 Model

Maritime surveillance relies significantly on efficient ship detection for various critical tasks, including vessel traffic management, search and rescue operations, and law enforcement. Traditional methods such as air surveillance, patrol vessels, and the Automatic Identification System (AIS) each have their own limitations. In this study, we investigate the effectiveness of deep learning based YOLOv8 in ship detection using Synthetic Aperture Radar (SAR) imagery, highlighting the potential of SAR as a complementary information source. In order to increase the effectiveness of the model’s learning, we created a balanced subset of the Large Scale SAR Ship Detection Dataset (LS-SSDD-v1.0), consisting of an equal number of ships and background classes. Despite the reduced dataset, our trained model shows comparable results, achieving 85.8% precision, 70.5% recall, and 77.6% mean Average Precision (mAP) on the imbalanced test set. Furthermore, we assessed the generalizability of the trained model by creating a custom dataset consisting of SAR imagery from the Gulf of Oman and the Persian Gulf. After fine-tuning for twenty epochs, the model achieved an accuracy of 90.4%