本文以南京城市为例 ,重点讨论了基于IKONOS影像的城市绿地信息分级分类提取方法 ,通过将IKONOS多光谱数据合成 ,根据各类地物的不同光谱特征 ,采取相应的方法提取出各层信息。在此过程中 ,仔细分析地物间在IKONOS 4个波段中的光谱差异 ,非线性增强阴影区绿地的NDVI值 ,利用光谱差异分层提取、剔除信息 ,最后把各分级绿地信息合并得到整体绿地分布图。分级分类法充分考虑各类目标的不同特点 ,避免了通常单一分类方法中单纯利用光谱特征所造成的地物混分现象。

This paper discusses about the extraction of urban green space from an IKONOS image using a hierarchical classification technique. Green space information was obtained based on the spectral characteristics of different objects with the help of available corresponding methods after the combination of IKONOS multi-spectral data. Due to high resolution of IKONOS imagery, large amount of data and heterogeneous nature of spectrum, the extraction of urban green space was carried out on segments after image segmentation. This would help much improving the accuracy of extraction of urban green space from the whole image. In test area of the image, the spectral characteristics of different features in all 4 bands are analyzed. The spectral characteristics of old urban area and asphalt road are similar to those of part of green space. Moreover, it is difficult to extract green space under the shadow. In order to extract information from the mixed green space with non-green space, through enhancing NDVI values of a green space under the shadow, parts of green space are extracted (NDVI > 0.18), then parts of non-green space are eliminated. The next step is to extract green space from mixed green space and non-green space based on spectral knowledge and unsupervised ISODATA clustering. Finally, green space information of test area is obtained by aggregating different levels of green space. The methodology is basically concerned with the object spectral features and noise due to the mixture of different land-use/land-cover categories is significantly avoided. To demonstrate the efficiency of proposed method, unsupervised ISODATA clustering method was used to extract green space from the test area,then both results were compared to show accuracy. The visual interpretation and ground truth checks of the test area have proved that the classification accuracy and productivity accuracy of the first method are higher than that of the latter.