高分辨率遥感影像中建筑物轮廓信息矢量化

发布时间：2018-03-06 22:29

本文选题：遥感影像　切入点：建筑物矢量化　出处：《遥感学报》2017年03期 　论文类型：期刊论文

【摘要】：针对高分辨率遥感影像的特点,提出了基于多类分割与模板匹配的建筑物轮廓矢量化方法:首先对影像进行多尺度SVM分割获取建筑物候选点;然后在Radon变换结合主轴分析获取建筑物主方向的基础上,引入多类分割思想,构建用于分割的能量函数,基于α-扩展算法解算能量函数,将轮廓线分割为3类边缘线段;接着构建形状先验的边缘模板,进行精确定位以获取边缘线的实际位置,去除锯齿状变形,降低提取结果的影响;最后相邻边缘线相正交得到拐角点,依次连接每一个拐角点,得到规则化的建筑物轮廓。相比于同类其他方法,此方法考虑了边缘点的方向信息和相邻边缘点趋于同一类的先验知识,可得到近似全局最优的边缘线段分割结果,避免了规则化过程中选择初始点和处理顺序的麻烦和不利影响,同时充分利用了影像特征,对边缘线段进行精确定位,减弱建筑物提取结果误差的影响。对不同影像的实验结果证明此方法可得到规则化的建筑物外轮廓线,不考虑提取过程中遗漏的建筑物,矢量化结果平均准确度为89%、完整度98%、几何形状相似性87%、整体质量85%。
[Abstract]:According to the characteristics of high resolution remote sensing image, a method of building contour vectorization based on multi-class segmentation and template matching is proposed. Firstly, multi-scale SVM segmentation is used to obtain the candidate points of the building. Then, on the basis of Radon transform and principal axis analysis to obtain the main direction of the building, the multi-class segmentation idea is introduced to construct the energy function for the segmentation. The energy function is solved based on the 伪 -expansion algorithm, and the contour is divided into three kinds of edge segments. Then the edge template of the shape prior is constructed to locate the actual position of the edge line accurately, remove the zigzag deformation, and reduce the influence of the extraction result. Finally, the corner points are obtained by orthogonal adjacent edge lines, and each corner point is connected in turn. Compared with other similar methods, this method takes into account the direction information of edge points and the priori knowledge that adjacent edge points tend to the same class, and obtains the approximate global optimal edge segment segmentation results. It avoids the trouble of choosing the initial point and processing order in the regularization process, and makes full use of the image features to locate the edge line segment accurately. The experimental results of different images show that this method can obtain the regular contour of the building without considering the buildings missing in the extraction process. The average accuracy of vectorization is 89, the integrity is 98, the geometric shape is similar to 87, the whole mass is 85.
【作者单位】：安徽省(水利部淮河水利委员会)水利科学研究院;安徽省大禹水利工程科技有限公司;河海大学地球科学与工程学院;清华大学地球系统科学研究中心;
【基金】：国家自然科学基金(编号:41271420/D010702)
【分类号】：TP751

【相似文献】