基于多重形态谱的高分辨率遥感影像舰船检测

Multiple Morphological Profiles for Ship Detection from High-Resolution Remotely Sensed Imagery

遥感影像的舰船检测在海洋运输、渔业监控和军事安全领域有着广泛的应用前景,因此受到了越来越多研究者的关注。而高分辨率的光学影像由于其图像内容丰富,是舰船检测中的一个重要数据源,但是糟糕的天气状况和影像中过多的细节信息都会干扰检测过程,增加检测难度。本文提出了一种基于多重形态学谱的舰船检测框架,快速而准确的从高分辨率的全色影像中提取舰船目标。该框架分为候选目标提取和舰船目标确认两个阶段。在候选目标提取阶段,本文设计了一种基于形态学的舰船指数,快速提取可能的舰船目标;在舰船目标确认阶段,首先提取了候选目标的多重形态谱特征,包括数学形态谱、属性形态谱和它们的差分形式,然后将它们输入随机森林分类器,剔除候选目标中的虚警。实验证明本文提出的框架可以快速准确的从高分辨率影像中提取舰船目标。

Ship detection from remotely sensed imagery has a wide range of applications in vessel traffic service, fisheries monitoring and military security, thus an increased number of researchers have paid attention to this field. High-resolution panchromatic imagery is an important data source for ship detection due to its abundant spatial information. However, the bad weather conditions and excessive details in high-resolution imagery can obstruct the detection. In this paper, we proposed a two-stage framework based on multiple morphological profiles to detect ships effectively. In candidate detection stage, a morphological ship index is built to detect ship candidates without any omissions. In candidate identification stage, structure features of the candidates are extracted from multiple morphological profiles, including morphological profiles and attribute profiles. A random forest classifier is subsequently employed to distinguish the true ships from false alarms. The experimental results show that the proposed framework achieves high detection accuracy in high-resolution optical imagery.