Maritime Object Detection based on YOLOx for Aviation Image

Accurate detection of ships in maritime scenarios is conducive to improving transport efficiency and reducing the occurrence of maritime traffic accidents. However, ships under the drone perspective are small and have various scale variations, affecting the detection algorithms. Aiming at this problem, this paper proposes a maritime object detection method based on YOLOx. First, the ship data in the maritime scenario is processed and screened to form a self-built dataset. Then, the retrained YOLOx model is used to detect ships in maritime scenarios. Finally, on the self-built dataset, CenterNet, YOLOv3, and YOLOv4 are used to conduct a comparative experiment with this method. Through the results of the comparative experiments, it is found that the detection accuracy of YOLOx is the best, reaching 90.86%. The method helps to promote the development of the application of drones in maritime scenarios.