Abstract: Ship detection algorithms applied to high-resolution optical remote sensing images are put forward. These algorithms are in order to solve the problem of high false alarm rate in traditional ship detection method in high-resolution optical remote sensing images. Two-dimensional image entropy which can describe the spatial texture feature is used to separate ocean and land combined with the principle of region growing. Visual saliency model is introduced in ship object segmentation. This method can solve the problem of unable to segment dark polarity ship target in ship target segmentation phase. The segmentation accuracy is higher in most background modes. Finally, in the segment candidate regions, multi-feature comprehensive method is adopted to eliminate false alarms. Results show that this method has higher detection rate and lower false alarm rate in ship target detection.

Key words: remote sensing image, object segmentation, ship detection, saliency, sea-land segmentation

摘要： 针对传统舰船检测方法在高分辨率光学遥感影像中虚警率较高的问题，提出了一种适用于高分辨率光学遥感影像的舰船检测算法。利用能够表征地物纹理特征的二维图像熵结合区域生长原理实现海陆分离，在舰船目标分割阶段，引入视觉显著性模型，解决了不能分割暗极性舰船目标的问题，大部分场景下分割精度较高。最后在分割出的候选目标中，采用多特征量综合的方法剔除虚警。结果表明，该算法在舰船目标检测中有较高的检测率和较低的虚警率。

关键词: 遥感影像, 目标分割, 舰船检测, 显著性, 海陆分离