对象置信度指引下的高分辨率遥感影像分割

发布时间：2018-03-08 12:10

  本文选题：高分辨率　切入点：遥感影像　出处：《仪器仪表学报》2017年09期 　论文类型：期刊论文

【摘要】：如何减小分割结果与实际地理对象间的差异,是目前高分辨遥感影像分割中面临的一个难点问题。为此,构建了一种新的对象置信度(OC)指标来衡量任意区域与地理对象间的匹配程度,进而提出了一种面向地理对象的多尺度分割算法。该算法主要包括两个步骤:首先,通过对影像进行过分割来构建初始种子区域集合,并确定尺度参数集合;而后,通过跟踪对象置信度指标OC的尺度间变化来指引多尺度区域合并过程,使区域合并结果逐步逼近实际的地理对象。多组实验结果表明,所提出的算法能够显著改善过分割及欠分割问题,准确识别建筑物、道路等地理对象的完整轮廓,在定性分析及定量精度评价中均显著优于商业软件e Congnition及传统多尺度分割算法。
[Abstract]:How to reduce the difference between segmentation results and actual geographic objects is a difficult problem in high-resolution remote sensing image segmentation. In this paper, a new object confidence index is constructed to measure the matching degree between any region and geographical object, and then a multi-scale segmentation algorithm for geographic object is proposed. The initial seed region set is constructed by over-segmentation of the image, and the set of scale parameters is determined. Then, the process of multi-scale region merging is guided by tracking the inter-scale variation of the confidence index OC of the object. The results of multiple experiments show that the proposed algorithm can significantly improve the over-segmentation and under-segmentation problems and accurately identify the complete contour of geographic objects such as buildings and roads. In qualitative analysis and quantitative accuracy evaluation, it is superior to commercial software e Congnition and traditional multi-scale segmentation algorithm.
【作者单位】： 南京信息工程大学电子与信息工程学院;河海大学计算机与信息学院;
【基金】：国家自然科学基金(61601229) 江苏省自然科学基金(BK20160966) 江苏省高校自然科学基金(16KJB510022) 东南大学移动通信国家重点实验室开放研究基金(2012D20)项目资助
【分类号】：TP751
，

本文编号：1583843