【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。更多还原

【Abstract】 How to reduce the difference between segmentation result and practical geographical object is a difficult problem faced by highresolution remote sensing image segmentation currently. Aiming at this issue,in this paper a new OC(Object Confidence) index is constructed to measure the matching degree between any region and geographical object,and a multi-scale segmentation algorithm facing to geographical objects is proposed. This algorithm mainly contains two steps: firstly,this algorithm establishes an initial seed regional set through conducting over segmentation to the image and determines the scale parameter set; secondly,this algorithm guides the process of multi-scale region merging through tracking the inter-scale change of OC index,and makes the region merging result gradually approach to practical geographical object. The multi-group experiments indicate that the proposed algorithm can obviously improve the over-segmentation and insufficient-segmentation problems,and identify the complete outlines of buildings,roads as well as other geographical objects accurately. The proposed algorithm is obviously superior to the commercial software e Congnition and traditional multi-scale segmentation algorithm in both qualitative analysis and quantitative precision evaluation.更多还原