基于超像素的高分辨率影像建筑物提取.docVIP

基于超像素的高分辨率影像建筑物提取 莫中夏张新长张志强 中山大学地理科学与规划学院 建筑物提取对于基础地理数据获取与更新具有重要意义。针对传统的基于像元方 法难以利用影像空间、上下文信息以及面向对象方法最优分割尺度难以确定等问 题，该文提出一种基于超像素的高分辨率影像建筑物提取方法。首先利用LiDAR 点云数据生成归一化高程模型以获取地物的高度特征，并通过Layerstacking 方式与高分辨率影像融合;随后基于Simple Linear Iterative Clustering (SLIC)算法分割高分辨率影像的RGB彩色图像牛.成超像素;继而基于分割所得 的超像素中的每个像元对超像素进行特征计算;最后使用基于RBF核的支持向量 机方法进行超像素级别分类，得到建筑物提取结果。实验结果表明，该方法能够 有效识别出简单场景和复杂场景下的建筑物，显著消除了传统基于像元方法出 现的“椒盐效应”，同时避免了面向对象方法中最佳分割尺度选择的难题，取 得了较好的建筑物提取效果。 关键词： 建筑物;超像素;SLIC; LiDAR; 基金：国家自然科学基金重点项0 ;国家自然科学基金面上项目 Building Extraction Based on Superpixels of High-Resolution Image MO Zhong-xia ZHANG Xin-chang ZHANG Zhi-qiang School ofGeography and Planning, Sun Yat—sen University; Abstract： Traditionally, the space and context information are difficult to be used in pixel-based method and the optimal segmentation scale is hard to be determined in the object-oriented method. To solve the problem, a high-resolution image building extraction method based on superpixel was proposed in this paper. Simple Linear Iterative Clustering (SLIC) is a kind of superpixels segmentation algorithm which has been developed in recent years. It has the following characteristics, which includes simple principle, fast segmentation, high degree of attachment to the boundary of the feature, and more attention to the shape. However, there are relatively few studies on the use of SLIC in the field of remote sensing. In this study, the RGB color image of the high-resolution image was segmented based on the SLIC algorithm to generate superpixels. Then, the normalized elevation model was generated by using the LiDAR point cloud data to obtain the height information of the features, and then it was merged with the high-resolution image by the way of layerstacking. Finally, the Support Vector Machine (SVM) method based on RBF kernel was used to classify the superpixels, and the result of building extraction was obtained. The experimental results showed that the proposed method could effectively ident