基于决策树的典型荒漠草地遥感分类策略

The classification strategy of desert grassland based on decision tree using remote sensing image

【目的】在遥感影像和地形图基础上对干旱半干旱地区天然荒漠草地进行分类,旨在为干旱半干旱地区草地分类及监测提供参考。【方法】采用基于决策树的天然草地分类方法,以天山北坡新疆阜康地区的TM遥感影像(2010-09-24)多光谱波段特征为主要分类变量,结合研究区数字高程模型(DEM),辅以影像目视解译结果,构建了对天然荒漠地区草地进行分类的决策树分类模型,对研究区主要地物类型进行了分类,并对分类精度进行了评价。【结果】综合考虑了遥感图像植被指数(NDVI)特征、DEM特征的分类策略,可以忽略山区的地形影响,简化分类过程,提高分类精度;与传统的非监督分类精度评价结果相比,基于专家知识的分类方法中总分类精度提高了37.4%,Kappa系数提高了78%,错分误差和漏分误差大幅减小;调整NDVI和DEM阈值可以使分类结果更加精准,模型适用区域更加广阔。【结论】基于决策树的典型荒漠草地遥感分类模型适用于荒漠区天然草地类型的提取与划分,可以提高利用遥感图像进行荒漠草地类型分类的精度。

[Objective] To improve grassland classification and monitoring in arid and semi-arid areas, this paper discusses the classification using remote sensing images and topographic maps. [Method] Decision tree was adopt to establish the classification model of natural grassland, supplemented by the Digital Elevation Model （DEM）, visual interpretation and the feature of multi-spectral band of TM remote sensing images in Fukang,Xinjiang （2010-09-24）. The feature extraction and classification of main surface types on the TM remote sensing images in northern Tianshan Mountains, was implement, then the results were evaluated. [Result] Considering the characteristic of normalized difference vegetation index （NDVI） from remote sensing image and the DEM feature, the classification process could ignore the impact of terrain in mountain, so as to simplify the classification and improve the accuracy. Compared with the traditional non-supervised classification method ISODATA, the total classification accuracy and Kappa coefficient of this method increased by 37.4% and 78%, respectively. Moreover, misclassification error was reduced significantly. Adjusting the NDVI and DEM threshold would make it more accurate and applicable. [Conclusion]This model is appropriate for the extraction and classification of natural desert grassland to improve the accuracy of remote sensing images for the classification of desert grassland types.