MoN-YOLO for Rotated Object Detection in SAR Ship Images

The development of deep learning provides a basis for detection techniques in SAR images. Nonetheless, challenges still remain when dealing with small and arbitrarily oriented ship objects. In this study, based on YOLOv8, we propose a rotated object detection algorithm called MoN-YOLO, which is specialized for ship detection in SAR images. The algorithm employs MobileNetV4 and introduces NwdLoss to optimize the detection bounding boxes. Experiments performed on the dataset (RSDD-SAR) showed that MoN-YOLO outperformed previous approaches, with a 1.6% increase in average precision, while also being computationally less expensive and having fewer parameters.