Rethinking Automatic Ship Wake Detection: State-of-the-Art CNN-Based Wake Detection via Optical Images

Most existing wake detection algorithms use Radon transform (RT) due to the long streak features of ship wakes in synthetic aperture radar images. The high false alarm rate of RT requires that the algorithms have significant human intervention for image preprocessing. When processing optical images, these algorithms are greatly challenged because it is not only sea clutter that interferes with the wake detection in the optical waveband but also many other environmental factors. Therefore, in this article, we address the automatic ship wake detection task in optical images from the idea of the convolutional neural network (CNN)-based object detection and design an end-to-end detector named WakeNet as a novel method. WakeNet processes all the wake textures clamped by the V-shaped Kelvin arms as the object and conducts detection via the oriented bounding box; the additional regression of wake tip coordinates and Kelvin arm direction can improve the hard wake detection performance while predicting the wake heading. According to the X-shaped wake spectral pattern and the wake edge enhancement principle in adjacent scales, spectral attention and multiscale attention modules are also used. Furthermore, more than 11.6k wake optical images are collected and annotated, providing a benchmark dataset. The experimental results on the dataset show that the proposed method has a better performance than other CNN-based methods, and the detection results of the yacht wake polarization images further verify its practical value. This article demonstrates that the deep learning-based method has considerable advantages over the traditional RT-based methods in wake detection.