基于视觉显著性的海面舰船检测技术  

Ship Detection on Sea Surface Based on Visual Saliency

作　　者：丁鹏[1,2] 张叶 贾平[1] 常旭岭[1] 刘让 

机构地区：[1]中国科学院长春光学精密机械与物理研究所中科院航空光学成像与测量重点实验室,吉林长春130033 [2]中国科学院大学,北京100049

出　　处：《电子学报》2018年第1期127-134,共8页Acta Electronica Sinica

基　　金：吉林省重大科技攻关专项(No.11ZDGG001);吉林省自然科学基金(No.20150101017JC)

摘　　要：海面舰船检测技术有着特殊的军用以及民用意义,为了在宽广且环境复杂多变的海面上快速、高效、精确地检测到舰船目标,本文提出了一种基于多特征、多尺度视觉显著性的海面舰船检测方法.该方法充分利用了四元数图像可在多个通道上同时进行操作,节省操作时间,并保证不同尺度特征之间关联性的特点;除此之外,该方法还利用人眼对不用大小的图像关注目标不同的特点对图像进行尺度大小变换以避免漏检.该方法先用顶帽算法对原图进行简单的图像预处理以抑制云层、油污的干扰;其次提取多种特征构成四元数图像进行舰船目标显著性检测;在得到显著图后利用OTSU分割算法确定舰船所在的区域,并在原图上标定、提取舰船目标.通过在多种海面情况下分别进行实验分析,实验结果表明该算法可以排除云、雾、油污等干扰,精确、快速地检测到舰船目标,真正率达96.52%,虚警率低至2.11%,相较于他显著性检测算法在舰船检测方面有明显的优势.The ship detection technology is of special significance in both military and civilian level. In order to detect ship targets quickly, efficiently and accurately in a wide and complex sea surface environment,a new method for ship detection based on multi-features and multi-scale visual saliency is proposed. This method makes full use of features of the hypercomplex images which can be operated simultaneously in a number of channels, save operation time, and guarantee the characteristics of different scale characteristics. First, the method uses top-hat algorithm for image preprocessing of the original image to suppress the interference of clouds and oil. Secondly,avariety of features are extracted to form hyper-complex images to detect the significance of ship targets. When we get the last saliency map,we segment ships to ensure the target location by using the OTSU algorithm, and then we mark the ship target in the original image. We make the experimental analysis in several sea conditions, and experimental results show that the algorithm can eliminate fog, cloud and grease interfering with accurate detection of ship targets. In this algorithm, true rate meets 96. 52% and the false alarm rate is as low as2. 11%. Compared to other saliency detection algorithm in ship detection, this algorithm has obvious advantages.

关 键 词：舰船检测 视觉显著性 PQFT算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]