基于WorldView-2影像城市绿地信息提取研究

WorldView-2Images Based Urban Green Space Information Extraction

以深圳市城市绿地为例,对遥感影像进行最佳波段组合、缨帽变换及主成分分析处理,并提取各地物在经处理后影像上的灰度值,利用决策树分类技术进行分类。结果表明:WorldView-2影像经过缨帽变换以及主成分分析处理后,能够明显增强影像的纹理信息,突出地物特征,并以各地物在经过缨帽变换及主成分分析处理的灰度值作为决策树分类的阈值,与以往的只利用个波段的灰度值及植被指数等作为阈值相比,精度明显提高,方法也得到改善,分类的总体精度、Kappa系数分别为90.36%、0.88,精度比较高,获得比较好的分类结果,但建设用地和耕地、绿地和耕地之间存在错分现象比较严重,主要是由于它们之间的纹理以及地物的光谱特征相似导致的。

In order to study the extraction of urban green space information based on WorldView-2remote sensing images,taking Shenzhen as an example,images of remote sensing were processed by optimal band combination,tasseled cap transform,and principal component analysis.The grey values of the ground objects were extracted from the processed images,and the decision tree technology was used for classification.The results showed that after dealing with the tasseled cap transform and principal component analysis,the image texture information was significantly enhanced with prominent terrain features.and the grey value could be used as the threshold value for decision tree classification,by which the degree of accuracy was improved significantly compared with those of using only grey value of a band and vegetation index as thresholds.The total accuracy was 90.36,and Kappa coefficient was 0.88,indicating the higher precision and better classification results of the new method.However,mistaken classification occurred seriously between the construction space and cultivated space,between green space and arable space,due to the texture and the similar features of the spectral characteristics.