基于Sentinel-2A时序数据和面向对象决策树方法的植被识别  

Identifying Vegetation with Decision Tree Model Based on Object-Oriented Method Using Multi-temporal Sentinel-2A Images

作　　者：毕恺艺 牛铮[1,2] 黄妮[1] 康峻[1,2] 裴杰[1,2] 

机构地区：[1]中国科学院遥感与数字地球研究所/遥感科学国家重点实验室,北京100101 [2]中国科学院大学,北京100049

出　　处：《地理与地理信息科学》2017年第5期16-20,27,128,共7页Geography and Geo-Information Science

基　　金：国家高技术研究发展计划(863)项目(2014AA06A511)

摘　　要：Sentinel-2A数据具有较高的空间分辨率和时间分辨率,克服了以往时序数据难以获取或空间分辨率低的问题。该文以山西省吕梁市陈家湾流域为研究区,基于Sentinel-2A时序数据,根据归一化植被指数(NDVI)时序曲线特征和光谱特征,构建基于面向对象决策树方法的分层分类模型,成功提取了陈家湾流域的植被信息,分类总体精度达到89.7%,Kappa系数为0.87。基于面向对象决策树方法的多时相分类结果与单时相分类结果相比,可以有效改善波谱特征相近和受地形影响较大地物的区分,减少混分现象;基于Sentinel-2A时序数据和面向对象决策树分类方法能够有效提高植被分类的精度。Time series satellite images play an important role in vegetation classification. However, most of the present resear-ches in using timt- series images are limited to low spatial resolution or the difficulty in obtaining remote sensed images. With the successful launch of satellite Sentinel- 2A, it s a good opportunity to construct NDVI time series w ith both high temporal and high spatial resolution to classify vegetation. In this paper, Chenijiawan watershed in Shanxi Province, China was chosen as the study area.The classfication was operated with a decision tree model that integrated objectoriented technology using Sentinel- 2A time- series images. This model chose appropriate spectral characteristics and seasonal characteristics to separate different types. The results indicated that Sentine- 2A satellite had a particular advantage in extracting information with its higher spatial and temporal resolutions, and its overall classification accuracy and the Kappa coefficient reached up to 89.7% and 0. 87, respec-tively.The results suggested that compared ith the vegetation identification results using a single temporal image, the results from time-series images had a better performance. The latter reduced the effect of terrain shadow and differentiated the objects with similar spectral characteristics. This study demonstrated that using mult-temporal Sentinel-2A data and decision tree mod＂ el based on object- oriented method can improve the vegetation classification accuracy.

关 键 词：Sentine-2A 时序数据 面向对象 归一化植被指数 决策树 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]