Ship Detection in Maritime Surveillance System Based on YOLOX

In this paper, we study the problem of ship target detection in maritime video surveillance systems, with special attention to the detection accuracy and adaptability under bad weather conditions. To this end, we propose an improved ship target detection method, including the addition of additive attention mechanism, and through testing on ship data sets, we verify the effectiveness of the proposed method. To further evaluate the performance of the method in severe weather conditions, we also used a flexible data enhancement strategy to generate a diverse synthetic test datasets, including different conditions such as rain, fog, and low illumination. Based on this datasets, we constructed a test datasets specifically for small vessel target detection, in which we evaluated the proposed approach. Experimental results show that compared with other mainstream competing algorithms such as Faster R-CNN, YOLOv3, YOLOv4, and YOLOX, our method has superior performance in terms of detection accuracy, robustness, and efficiency. In addition, the proposed method proves its good adaptability in ship detection under harsh imaging conditions. Therefore, the improved method proposed in this paper is expected to provide higher accuracy and reliability for ship target detection in Marine video surveillance systems.