Bi Fukun, Chen Jing, Zhuang Yin, Bian Mingming, Zhang Qingjun
Department of Electronic and Information Engineering, North China University of Technology, Beijing 100144, China.
Department of Information and Electronic, Beijing Institute of Technology, Beijing 100081, China.
Sensors (Basel). 2017 Jun 22;17(7):1470. doi: 10.3390/s17071470.
With the rapid development of optical remote sensing satellites, ship detection and identification based on large-scale remote sensing images has become a significant maritime research topic. Compared with traditional ocean-going vessel detection, inshore ship detection has received increasing attention in harbor dynamic surveillance and maritime management. However, because the harbor environment is complex, gray information and texture features between docked ships and their connected dock regions are indistinguishable, most of the popular detection methods are limited by their calculation efficiency and detection accuracy. In this paper, a novel hierarchical method that combines an efficient candidate scanning strategy and an accurate candidate identification mixture model is presented for inshore ship detection in complex harbor areas. First, in the candidate region extraction phase, an omnidirectional intersected two-dimension scanning (OITDS) strategy is designed to rapidly extract candidate regions from the land-water segmented images. In the candidate region identification phase, a decision mixture model (DMM) is proposed to identify real ships from candidate objects. Specifically, to improve the robustness regarding the diversity of ships, a deformable part model (DPM) was employed to train a key part sub-model and a whole ship sub-model. Furthermore, to improve the identification accuracy, a surrounding correlation context sub-model is built. Finally, to increase the accuracy of candidate region identification, these three sub-models are integrated into the proposed DMM. Experiments were performed on numerous large-scale harbor remote sensing images, and the results showed that the proposed method has high detection accuracy and rapid computational efficiency.
随着光学遥感卫星的快速发展，基于大规模遥感图像的船舶检测与识别已成为重要的海洋研究课题。与传统远洋船舶检测相比，近岸船舶检测在港口动态监测和海事管理中受到越来越多的关注。然而，由于港口环境复杂，停靠船舶与其相连的码头区域之间的灰度信息和纹理特征难以区分，大多数常用检测方法受到计算效率和检测精度的限制。本文提出了一种新颖的分层方法，该方法结合了高效的候选区域扫描策略和精确的候选区域识别混合模型，用于复杂港口区域的近岸船舶检测。首先，在候选区域提取阶段，设计了一种全向二维交叉扫描（OITDS）策略，用于从水陆分割图像中快速提取候选区域。在候选区域识别阶段，提出了一种决策混合模型（DMM），用于从候选目标中识别真实船舶。具体来说，为提高对船舶多样性的鲁棒性，采用可变形部件模型（DPM）训练关键部件子模型和整船子模型。此外，为提高识别精度，构建了周围相关上下文子模型。最后，为提高候选区域识别的准确性，将这三个子模型集成到所提出的DMM中。在大量大规模港口遥感图像上进行了实验，结果表明所提方法具有较高的检测精度和快速的计算效率。
