YOLOv8-Sea: An Efficient Object Detection Network for Sea Surface Scenarios

Object detection in sea surface scenarios is a key task for unmanned surface vehicles (UAVs) to achieve autonomous navigation. This task usually needs to overcome difficulties such as different target scales, complex sea surface background and small targets at a long distance. To address these challenges, this paper proposes YOLOv8-Sea, a sea surface scene-oriented object detection network based on YOLOv8s. Our network introduces the Star-c2f module to improve the network's high-dimensional and nonlinear feature transformation capabilities; the Global-to-Local Spatial Aggregation (GLSA) module to enhance the network's ability in extracting object features from complex backgrounds; and the introduction of Inner-IoU loss to achieve faster and more efficient regression loss. By conducting experiments on a self-made sea surface object detection dataset, the experimental results show that our proposed network is 3.4% more accurate than the original YOLOv8s, and [email protected]:95 is 2.5% better than the baseline model.