【摘要】 为了高效准确地基于“珠海一号”高光谱遥感数据对油茶(Camellia oleifera)林进行识别提取,本研究以湖南省邵阳县小溪市乡作为研究区,“珠海一号”高光谱影像作为数据源,ASTER GDEM数字高程数据、Google Earth高分辨率数据作为辅助数据,借鉴决策树分层提取思想,采用一种光谱、纹理以及地形特征相结合的分级提取方法提取油茶林。油茶提取体系分为3级,首先通过不同特征结合提取新油茶林与非植被覆盖地;在此基础上对草地与耕地同样采用多特征结合进行提取;最后针对林地构建2个油茶植被指数提取老油茶林。结果表明:基于高光谱数据多特征结合进行分层提取的总体分类精度达到86.82%,Kappa系数为0.83,而使用随机森林进行分类的总体精度仅是82.15%,Kappa系数为0.78。研究证明基于高光谱遥感数据的多特征结合分层提取方法能有效识别并提取油茶林,可为今后油茶林以及其他土地利用类型提取提供参考。更多还原

【Abstract】 In order to identify and extract Camellia oleifera plantations from “Zhuhai-1” hyperspectral remote sensing data efficiently and accurately,this study proposed a hierarchical extraction method based on multi-feature combination.Taking Xiaoxishi Towship of Shaoyang County,Hunan Province as the research area,“Zhuhai-1” hyperspectral images,ASTER GDEM digital elevation data were used as data sources,a hierarchical extraction method combining spectrum,texture and terrain features was adopted to extract C.oleifera plantations by referring to the hierarchical extraction idea of decision tree.The extraction system was divided into three levels.Firstly,the new C.oleifera plantation and non vegetation covered land were extracted through the combination of different characteristics.The grassland and cultivated land were then extracted by feature extraction.Finally,two vegetation indices of C.oleifera were constructed to extract the old C.oleifera plantation,so as to achieve the purpose of extracting C.oleifera plantation.The results showed that the overall classification accuracy of hierarchical extraction based on multi-feature combination of hyperspectral data was 86.82%,and the Kappa coefficient was 0.83,while the overall accuracy of classification was 82.15%,and the Kappa coefficient was 0.78 by using by random forest method.This study proves that the multi-feature combined with hierarchical extraction method based on hyperspectral remote sensing data can effectively identify and extract C.oleifera plantation,and provides a reference for the extraction of C.oleifera plantation and other land use types in the future.更多还原