摘要
遥感图像空间分辨率的提高,为目标物的纹理特征和形状特征的提取提供了客观基础,同时也使得传统的基于像元的分类识别方法受到了严重的挑战。因此,需要对传统的方法进行改进或发展新的方法。本文采用面向对象的分析思想,通过图像分割和分割对象的矢量化等一系列的预处理,并在此基础上实现了目标形状信息的提取,最后综合利用光谱特征和形状特征应用模糊分类器实现两种典型的人造目标的分类提取实验。识别的精度评价主要通过目视解译完成。分析表明,形状信息的提取大大丰富了目标识别的特征库,尤其在感兴趣目标与背景物具有相近的光谱反应而形状特征有明显差异的条件下,这种利用光谱与形状特征整合的提取方法能够大大提高目标的识别精度。
With the improvement of the spatial resolution of remote sensing image, the objective basis is provided for extracting the texture and shape features, at the same time, the traditional pixel-based classification methods are challenged severely. So it is necessary to improve existing methods or to develop new one. In this paper, according to object-oriented analysis method, firstly a serial of pre-processing procedures are performed, such as image segmentation, edge tracing and vectorization, and vectorization compression; then the shape features are extracted from the vectorization information, finally with the help of the spectral feature and shape features, the classification for two kinds of typical artificial objects is finished by using the fuzzy classifier, and the classification accuracy is evaluated by visual interpretation. The results show that the extraction of shape features enriches enormously the feature database for object identification, especially under the condition when the object of interest and background have the similar spectral reflection and the apparent different shape features, this object-oriented classification by integrating spectral and shape features can improve greatly the identification accuracy.
出处
《遥感学报》
EI
CSCD
北大核心
2006年第1期104-110,共7页
NATIONAL REMOTE SENSING BULLETIN
基金
中国科学院知识创新方向性项目"定量遥感应用的几个关键问题研究"(KZCX3-SW-338-4)
国家自然基金(30070604)
关键词
基于对象
分割
矢量化
形状特征
目标识别
object-oriented identification
segmentation
vectorization
shape features
object identification
