基于滑动窗口的遥感图像人造目标检测算法  

Man-made Object Extraction from the Remote Sensing Image Based on Sliding Windows

作　　者：李玲玲[1] 刘永进 王自桦 金泰松[2,3] 

机构地区：[1]郑州航空工业管理学院计算机科学与应用系,河南郑州450015 [2]光电控制技术重点实验室,河南洛阳471004 [3]厦门大学信息科学与技术学院,福建厦门361005

出　　处：《厦门大学学报（自然科学版）》2014年第6期792-796,共5页Journal of Xiamen University：Natural Science

基　　金：国家自然科学基金(41171341);河南省重点科技攻关计划项目(132102210054);福建省自然科学基金(2011J01365);航空科学基金(20125168001)

摘　　要：提出一种基于滑动窗口的人造目标检测算法.首先,根据图像的平滑性将整个图像划分为平滑与非平滑图像区域;然后,在非平滑区域随机生成一系列滑动窗口,并融合4种目标特征的度量准则构造一个判别滑动窗口目标性的评分函数,检测算法根据评分函数的取值判别窗口中是否包含人造目标;最后,在人造目标图像集上进行的实验表明:该算法能准确地检测出图像中的人造目标,并具有一定的抗噪声干扰能力.An approach to extract man-made objects from the remote sensing image is proposed.First,the whole image is classified into smooth morphology regions and non-smooth morphology regions;second,the set of different sizes of sliding windows is randomly generated in the non-smooth image regions,presenting a measure to quantify how likely it is for the sliding window to contain the man-made object.The proposed measure combines general object measure and the man-made object measure to present an evaluation function;and the man-made object regions of the remote sensing image are obtained based on the evaluation function value of different sliding windows;Finally,the object extraction test on the image dataset demonstrates that the proposed algorithm can correctly locate sliding windows which contain the man-made object and at the same time,it is resistant to noises.

关 键 词：人造目标 分形 纹理 检测 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]