基于面向对象和无人机影像的黄土高原丘陵区小流域梯田提取研究  

Extraction of Small Watershed Terraces in the Hilly Areas of Loess Plateau Through UAV Images with Object-oriented Approach

作　　者：张艳超 杨海龙[1] 信忠保[1] 鹿琳琳 ZHANG Yanchao;YANG Hailong;XIN Zhongbao;LU Linlin(College of Soil and Water Conservation,Beijing Forestry University,Beijing 100083;Aerospace Information Research Institute,Chinese Academy of Sciences,Beijing 100094)

机构地区：[1]北京林业大学水土保持学院,北京100083 [2]中国科学院空天信息创新研究院,北京100094

出　　处：《水土保持学报》2023年第3期139-146,共8页Journal of Soil and Water Conservation

基　　金：国家自然科学基金项目(42177319,41877539)。

摘　　要：精确提取梯田对于水土保持动态监测和评价非常必要,建立黄土高原小流域梯田提取技术流程可为水土流失防治提供技术支持。以甘肃省天水市水土保持科学试验站坚家山小流域为研究区域,采用无人机影像数据,基于尺度参数评估工具以影像的纹理特征为输入层,确定最优分割尺度参数;选用光谱特征和可见光植被指数为分类特征,使用面向对象分类方法对梯田进行提取。结果表明:(1)以局部方差变化率101作为多尺度分割的尺度参数时,梯田边界明显,光谱、纹理和形状特征对于描绘梯田边界具有巨大的潜力;(2)采用可见光植被指数植被颜色指数(CIVE)、超绿指数(EXG)、超绿超红差分指数(EXGR)、绿蓝比值指数(GBRI)、绿红比值指数(GRRI)、归一化蓝绿差异指数(NGBDI)、可见光波段差异植被指数(VDVI)、Woebbecke指数(WI)分别作为梯田提取的分类特征,其中,EXG植被指数精度最高,梯田提取精度为72.60%,并提出一种基于最邻近分类器以综合指数(CIVE、WI、EXG、EXGR)为分类特征,建立分类阈值的梯田提取方法,梯田提取精度为91.20%,相较于以单植被指数的分类方法精度提高18.60%。研究基于无人机影像的多尺度分割的面向对象方法可适用黄土丘陵区梯田的提取,尤其采用综合植被指数可以显著提高分类精度。Extracting terraces accurately is very necessary for soil and water conservation monitoring and establishing a method of terraced field extraction in small watershed could provide technical support for soil erosion prevention and control on Loess Plateau.Taking Jianjiashan sub-watershed of Tianshui City,Gansu Province Water and Soil Conservation Scientific Experiment Station as the study area,this study tries to use object-oriented classification method to extract terraces by using Unmanned Aerial Vehicle(UAV)image data,which works by determining the optimal segmentation scale parameters based on the scale parameter evaluation tool with texture features of images as the input layer and selecting spectral features and visible vegetation index as the classification features.The results showed that:(1)when the local variance change rate is 101,the terraces boundary is obvious,which shows that the spectral,texture and shape features have great potential for depicting the boundaries of the terraces;(2)the single vegetation indices Color index of vegetation(CIVE),Excess green(EXG),Excess green minus excess red(EXGR),Green/Blue reflectance ratio(GBRI),Green/Red reflectance ratio(GRRI),Normalized green-blue difference(NGBDI),Visible difference vegetation index(VDVI)and Woebbecke(WI)were used as the classification features for terraces extraction,respectively.Among them,EXG vegetation index had the highest accuracy on terraces extraction,with an accuracy of 72.60%.And a method of terrace extraction based on nearest classifier was proposed in this study,which took the composite index(CIVE,WI,EXG,and EXGR)as classification feature to established classification threshold.The accuracy of new method proposed in this study was 91.20%,which was 18.60% higher than that of the classification method with single vegetation index.The object-oriented method based on multi-scale segmentation of UAV images can be applied to the extraction of terraces in loess hilly area,and the comprehensive vegetation index can significantly improve th

关 键 词：梯田 UAV 多尺度分割 面向对象 

分 类 号：S127[农业科学—农业基础科学] S157.31