一种海上弱小运动船舶实时检测方法

A real-time detection method for dim and small moving ships at sea

弱小船舶目标实时检测因在海上搜救、无人船和海上交通管理等领域中的众多应用而备受关注。虽然基于深度学习的目标检测算法,如YOLO(you only look once)和SSD(single shot multibox detector)等取得了不错的目标检测性能,但是它们仍然无法实时有效检测出海上弱小船舶运动目标。针对此问题,文章提出了一种改进的深度学习网络结构,结合SELU(scaled exponential linear units)激活函数,有效解决了已有的YOLOv2算法对弱小目标检测率较低的不足以及YOLOv3算法中残差网络结构冗余的问题。实验表明,该文提出的方法在海上弱小船舶目标检测上,比原YOLO算法具有更高的检测精度、更快的检测速度和更优良的鲁棒性。该方法在低配硬件环境中仍具有实时性的特点,因此对算法的推广应用具有实际的意义。

Real-time detection of dim and small ships has attracted more and more attention in search and rescue on sea,unmanned surface vehicle(USV),and ocean traffic management recently.Deep learning based target detection algorithms,such as you only look once(YOLO)and single shot multibox detector(SSD),have achieved good target detection performance.However,it is still difficult for them to effectively detect dim and small ship targets at sea in real time.In this paper,an improved network structure based on deep learning is proposed to solve the problems.Integrated with scaled exponential linear units(SELU)activation function,it improves the ability of YOLOv2 algorithm in detecting dim and small targets,and diminishes the redundancy of residual network structure in YOLOv3 algorithm.Experimental results show that the improved algorithm has higher detection accuracy and speed,and more robustness than the original YOLO algorithms in detecting the dim and small ship targets at sea.And its real-time detection based on low configured hardware makes the algorithm more applicable in reality.