A Directional and Multiscale Features-Based Method to Extract Ship Wake in SAR Images

Abstract

Ship wake is of great significance for ship positioning and navigation situation estimation using remote sensing images. However, the similarity between ship wakes and the complex background of the ocean surface leads to low detection accuracy. Furthermore, the current ship wake detection methods are mostly based on simulated SAR images, which are insufficiently applicable to wake detection in real SAR images. Herein, we propose a YOLOv8s-OBB-based ship wake detection network YOLO-RealSW. We adopt the dynamic directional convolution module to extract multidirectional distribution features through directional-aware convolution, and the spatial combined attention mechanism to suppress sea clutter and noise interference, thereby enhancing feature discriminability. We also proposed a directional convolution wavelet transform module to strengthen the wake structure information in the frequency domain and enhance the wake features. Experiments on the public dataset OpenSARWake show that this method is superior to the existing mainstream methods in terms of accuracy and robustness. In addition, we use the SARWakeTest test set based on Sentinel-1A to test in different real scenes and sea conditions, which provides an effective solution for ship wake detection tasks in various scenarios.