基于面向对象信息提取技术的城市用地分类

针对高分辨率遥感影像的城市用地分类,引入了面向对象的信息提取技术,并将其与传统基于像素光谱信息的分类方法进行了比较。在此基础上详述了面向对象信息提取的关键技术---多尺度影像分割和基于分割的分类技术。以城市作为研究区,实现城市用地的自动分类。图像处理过程包括几何校正、HIS融合、图像分割和图像分类。最终分类结果表明:视觉上,面向对象信息提取技术克服了传统方法无法克服的“椒盐”噪声的影响;精度上,面向对象信息提取技术的总体精度高达84.82%,比最大似然法的总体精度提高了10.95%,并且各类地物信息的提取精度均有所提高,其中草地、道路、建筑物阴影的精度较高。

Classification of Urban Land Based on Object-oriented Information Extraction Technology

Object-oriented information extraction technology compared with the pixel-based classification method is suitable for classification of high resolution remotely sensed images. Object-oriented image analysis has two key technologies, multi-scale image segmentation and classification technologies based segmentation. Urban area of Huairou was selected as study area, and the purpose is to extract information from QuickBird image using above approach. The conclusions are:①“pepper and salt”noises are discarded;②84.82% overall accuracy is achieved while only 73.87% is achieved with traditional pixel-based method. Furthermore, precision of each kind of object information was also improved, particularly for grass, roads and building shadows.