摘要
为快速获取农情信息与农作物种植结构,通过面向对象的识别方法,对新疆主要粮食产区之一奇台县进行作物信息提取的研究。以Landsat 8遥感影像为数据源,通过更新2016年土地利用现状图得到耕地分布信息。使用e Cognition 9. 0软件进行多尺度分割,通过ESP2插件确定研究区最佳分割尺度后进行尺度分割,结合实地调查资料,利用面向对象的Cart决策树分类器和随机森林分类器将作物分为小麦、玉米、打瓜和葵花四类主要作物,提取新疆奇台县作物种植信息。结果表明:研究区最佳分割尺度为90;对于本研究,随机森林分类器Cart树数量为80~90时分类精度较高;Cart决策树总体精度达到0. 925,Kappa系数0. 893;随机森林分类器总体精度达到0. 945,Kappa系数0. 921。表明,在县域级农作物识别时使用面向对象的识别方法对中等空间分辨率遥感影像分类是可行的。
In order to obtain agricultural information and crop planting structure quickly,the extraction of crop information in Qitai County as one of the main grain producing areas of Xinjiang was studied by object-oriented identification method.The remote sensing images of Landsat 8 were used as data sources,and the cultivated land distribution information was obtained by updating the present landuse map in 2016.Based on the optimal segmentation scale determined by ESP2,the eCognition 9.0 software was used for multi-scale segmentation of farmland in Qitai County.Combined with the field survey data,four main crops including wheat,corn,melon and sunflower were divided and their planting information were extracted by object-oriented Cart decision tree classifier and random forest classifier.The results showed that the optimal segmentation scale for the research area was 90.The classification accuracy was higher when the number of Cart tree in the random forest classifier was 80~90.The overall precision of Cart decision tree classifier was 0.925 and the Kappa coefficient was 0.893,while those of the random forest classifier were 0.945 and 0.921,respectively.Therefore,it was feasible to use the object-oriented recognition method to recognize the remote sensing images of crops at the county level from medium spatial resolution.
作者
吕昱
范燕敏
武红旗
彭田田
皇甫蓓炯
贺梦婕
LU Yu;Fan Yanmin;Wu Hongqi;Peng Tiantian;Huangfu Beijiong;He Mengjie(College of Grassland and Environment Sciences,Xinjiang Agricultural University,Urumqi 830052,China;Qitai County Agricultural Technology Extension Center,Qitai 831800,China)
出处
《山东农业科学》
2020年第6期137-143,共7页
Shandong Agricultural Sciences
基金
国家自然科学基金项目“基于无人机平台的滴灌棉花规模化种植过程中营养快速诊断指标及建模研究”(31560340)。
