基于面向对象的干旱半干旱地区植被分类

Vegetation classification in arid and semi-arid areas using an object-oriented method

为了提高干旱半干旱地区不同植被的分类精度,以多尺度分割后的Sentinel-2A影像为主要数据源。通过融合主成分变换分析、植被指数以及影像纹理特征等,对比分析了CART决策树、C4.5决策树、KNN、SVM 4种分类模型在干旱半干旱地区面向对象的分类精度。结果表明:面向对象分类的最佳分割尺度为58、81和102,即在102尺度下分离出植被和非植被后,分别在58、81尺度下提取不同植被的特征信息。由分类精度可知,基于决策树的分类精度高于KNN、SVM算法,各模型的分类精度均达到89%以上,其中CART决策树分类总体精度最高达到91.28%,Kappa系数0.91,验证了基于中分辨率单时相遥感影像进行复杂下垫面植被识别的可行性。

In order to improve the classification accuracy of different vegetation types in arid and semi-arid areas,a Sentinel-2A image was used for classification in this paper after multi-scale segmentation.Using principal component transformation analysis,vegetation index construction,and image texture feature extraction,the object-oriented classification accuracy of the CART decision tree,C4.5 decision tree,KNN,and SVM in the south of Horqin sandy land was compared and analyzed.The optimal segmentation scales of object-oriented classification were 58,81,and 102,and after separating vegetation and non-vegetation at the 102 scale,characteristic information on different vegetation types was extracted at the scales of 58 and 81.The classification accuracy of the decision tree was higher than that of the KNN and SVM algorithms,and the classification accuracy of all models exceeded 89%.The overall classification accuracy of the CART decision tree was as high as 91.28%,and the kappa coefficient was 0.91.The feasibility and applicability of vegetation identification based on moderate-resolution single time-phase remote sensing images were verified in this study.