【作者】 李海月；

【导师】 王宏琦；

【作者基本信息】 中国科学院研究生院（电子学研究所） ， 信号与信息处理， 2006， 硕士

【摘要】 自动目标识别(Automatic Target Recognition，ATR)技术，是遥感图像处理领域的核心技术之一。近年来，随着几颗商用遥感卫星的成功发射，如IKONOS、QuickBird等，高分辨率遥感图像资源越来越丰富，遥感图像中的自动目标识别技术再次成为研究的热点。遥感图像中自动目标识别的任务是将感兴趣的目标，如森林、湖泊、农作物、建筑物等，从图像背景中自动地识别出来。为了实现这一目标，需要综合应用图像处理，特征提取，图像分割等多项技术。针对实际的应用，在完成目标的自动识别后，还需要对其进行几何重建，构造出目标的规则几何外形，以实现地图的自动标绘。遥感图像中的自动目标识别与标绘技术现在已广泛应用于环境监测、军事侦察、农业估产、灾害控制等方面，也是未来地理信息系统(GIS)、数字地球、数字城市构想的关键技术。 本文研究了当前遥感图像中建筑物自动识别与标绘领域的相关技术和方法。在对已有方法进行归纳总结的基础上，提出了一种建筑物自动识别与标绘的应用框架。该框架包含三个部分：首先从城区遥感图像中找出建筑物的候选区域，然后从这些候选区域中确定真实的建筑物目标，最后对建筑物进行标绘。 在建筑物候选区域的寻找阶段，本文主要采用了一种基于灰度共生矩阵的方法对图像进行分割，从分割结果中得到建筑物的候选区域。 在建筑物目标的确认阶段，本文提出了一种基于主线段的建筑物判别方法。该方法通过区域生长、边缘提取、Hough变换求取建筑物候选区域的主线段，然后通过主线段长度来判别该候选区域是否为真实的建筑物目标。 在建筑物外形标绘阶段，本文提出了一种网格匹配与形态学处理相结合的方法，可获得建筑物的规则几何外形。 本文所进行的各项实验，均采用近年来QuickBird所采集的世界各大城市的高分辨率图像。实验结果表明，本文提出的应用框架对城区建筑物目标进行识别和标绘，有较好的准确性和鲁棒性。更多还原

【Abstract】 Automatic Target Recognition (ATR) is a core technology in the field of remote sensing image processing. Recent commercial satellites, like IKONOS and QuickBird, provide a large quantity of high resolution remote sensing images. The ATR technology has become more useful and important. The goal of ATR in remote sensing images is to recognize the regions of interest, such as forests, lakes, crops, and buildings. In order to achieve this goal, multiple techniques like image processing, image feature extraction, image segmentation are used. After a target is recognized, a mapping procedure is needed to acquire the regular outline of this target. Automatic target recognition and mapping techniques are widely used in environment monitoring, military detective, agricultural estimation, and disaster control. They will be the key techniques of future GIS, digital globe, digital city system.The researches of this paper focus on automatic building recognition and mapping techniques in urban remote sensing images. By concluding and summarizing former methods, this paper proposes an application framework for automatic building recognition and mapping. There are three steps in this framework: candidate building area searching, building target verification, and building mapping. In the candidate building area searching step, an image segmentation method based on grey level co-occurrence matrix (GLCM) is adopted to get candidate building areas in an urban image.In the building target verification step, a method based on the dominant line segment is proposed. The dominant line segment (the longest line segment) of every candidate area is extracted by a procedure using region growing, edge detection, and Hough transform. Then, the real buildings are verified by judging the length of their dominant line segments.更多还原