Improved YOLOv8-OBB Model for Ship Wakes Detection in SAR Imagery

Ship wakes exhibit significant features on the synthetic aperture radar (SAR) images, which have been an important way to identify characteristics of ship type, position, speed and heading information. Compared with the ship hull, it is difficult to hide and fast fade.However, the high dynamics of the marine environment under complex sea state brings challenges to detecte wakes.This paper presented an improved method for wakes detection with YOLOv8 framework to solve the problem of weak features and sea clutter interference in complex sea states. The proposed algorithm exploits the rotating target detection capability of YOLOv8-OBB, by introducing the SE Attention mechanism to recalibrate the channel weights and amplify the wake-up feature responses. Meanwhile, to reduce the redundant computation and improve the feature fusion efficiency, a lightweight CSPPC convolution module is adoptsed instead of the standard convolution module. The effectiveness of the modual has been test performed on the OpenSARWake dataset and ablation experiments. The results demonstrate that the YOLOv8-OBB baseline has improved 0.8% accuracy than YOLOv11-OBB in mAP, but the model size is 0.93 Mb larger. The improved model in this paper has a notable improvement with 2.4% improvement in mAP and 0.71 Mb reduction in size compared with baseline,which achieves the model lightweight and improved detection accuracy. In addition,, the improved model performs better capability than the baseline and YOLOv11-OBB against complicated sea state conditions based on the constructed SARWake_Seastate dataset, which further verifies the robustness of the model in this paper.