辅以NDVI/DEM的面向对象木薯提取方法研究——以广西壮族自治区武鸣县为例

Research on Object-Oriented Classification Method Assisted with NDVI/DEM in Extracting Cassava:Taking Wuming County for Example

木薯作为重要的非粮能源作物,因其种植分散、与易混淆作物缺乏生长时相差,从而导致其种植分布信息难以正确获取,一直是困扰木薯乙醇资源正确评估的技术问题。该研究以广西壮族自治区武鸣县为研究区,应用高分辨率RapidEye影像数据,探讨利用面向对象分类方法合理提取木薯种植面积及其空间分布信息。研究表明,将归一化植被指数(NDVI)和数字高程数据(DEM)应用于遥感影像的多尺度分割,并结合基于隶属度函数和阈值的面向对象分类方法,提取木薯种植面积的精度达85%,分类精度(以Kappa系数表示)为0.9。相比最大似然监督分类方法和未辅以NDVI/DEM的面向对象分类方法,该方法的总精度分别提高了5%和12%,Kappa系数分别提高了0.2和0.3。因此,NDVI和DEM数据参与影像分割的面向对象的信息提取方法,可以有效地提高遥感图像分类的精度,并为提取种植分散、与相关植被时相差异小的作物的空间分布提供了有效的技术借鉴。

Cassava is an important feedstock not only for animal feeding but also for bio-fuel. It has a significant meaning to make it clear where the cassava grows and how much it is. But it has been the technical obstacles to assess the resource of cassa- va accurately that the cassava is dispersive and has no obvious growth differences with other corps easy to be confused. In this research, taking Wuming County of Guangxi Autonomous Region as the study area, an object-oriented image analysis method to extract the spatial distribution area of cassava was explored based on the RapidEye images with high resolution by eCognition 8. 7, ERDAS9.3 and ArcGIS10. 1. It is shown in the study that： 1）Applying NDVI and DEM data in the multi-scale segmenta- tion and based on membership function and threshold value, the object-oriented classification method has the precision of 85M and the Kappa coefficient of 0. 9 for the extraction of cassava distribution area. 2）Compared with the maximum-likelihood classi- fication method and the object-oriented classification method without NDVI/DEM segmentation, the object-oriented classifica- tion method with NDVI/DEM segmentation in this paper has higher classification accuracy for the extraction of cassava, with the precision higher 5% and 12% and the Kappa coefficient higher 0. 2 and 0. 3 than that of the former two respectively. There- fore, the object-oriented classification method with adding NDVI and DEM data to the image segmentation can improved effi- ciently the classification accuracy of remote sensing image, and it can provide an effective technical reference for the extraction of crops with dispersed planting area and having little time differences with other confusing plants.