GOAL-ORIENTED SEMANTIC MODULES FOR SAR SHIP DETECTION

Synthetic Aperture Radar (SAR) imaging is a vital technology for maritime surveillance, enabling ship detection under all weather conditions and lighting scenarios. Recent advancements in Deep Learning (DL) have substantially improved the performance of SAR Ship Detection systems. Among these advancements, various adaptations of the You Only Look Once (YOLO) model have been developed to address the unique challenges of SAR imaging, such as varying ship sizes, complex backgrounds, and high noise levels. This paper reviews and analyzes studies incorporating innovative techniques like Multi-Scale Feature Extraction (MSFE), Attention Mechanisms, and Lightweight Neural Networks (LNN) to enhance detection accuracy and efficiency. Through a comparative analysis, we evaluate the strengths and weaknesses of these adaptations across different datasets. Additionally, we present the implementation of a modified YOLOv5 Small (YOLOv5s) model on the Pixel-wise Segmentation SAR Ship Detection Dataset (PSeg-SSDD), achieving a balance between accuracy and computational efficiency. Our findings demonstrate the model’s suitability for advanced maritime monitoring applications.