High-Speed Ship Detection in SAR Images by Improved Yolov3

Abstract

Ship detection in synthetic aperture radar (SAR) images plays an important role in marine transportation, fishery management, and maritime disaster rescue. Nowadays, the current researches almost are focusing on improving detection accuracy while detection speed is neglected. However, it is also extraordinarily important to increase the ship detection speed, because it can provide real-time ocean observation and timely ship rescue. Therefore, in order to solve this problem, this paper proposes a high-speed SAR ship detection approach by improved you only look once version 3 (YOLOv3). We experimented on a public SAR ship detection dataset (SSDD) which has been used by many other scholars. Finally, the experimental results indicated that the detection speed of our proposed improved YOLOv3 is faster than current other methods, such as faster-regions convolutional neural network (Faster R-CNN), single shot multi-box detector (SSD), and original YOLOv3 under a same hardware environment. Meanwhile, the detection accuracy remains basically unchanged.