OS-Net: A novel oriented ship detector based on RetinaNet

Ship inspection plays an important role in maritime traffic management and the maintenance of national maritime security. However, the different bow orientations and dense arrangement of vessels make horizontal frame detection inappropriate for ship detection. Therefore, in this paper, a rotating frame detection network based on RetinaNet [1], named OS-Net, is proposed. In order to cope with the situation of obscured vessels such as clouds, the data enhancement strategy represented by Cutout [2] is used in this paper to improve the network generalization and learn to cope with the situation of obscured pictures. In order to fully extract the ship features to improve the detection accuracy, this paper adopts the ConvNext [3] network as the backbone network and explores the improvement effect on the detection accuracy by deepening the ConvNext depth. The well-known HRSC2016 dataset is used in the paper, and through a large number of ablation experiments, we can find that the detection accuracy of OS-Net is more excellent.