摘要
卫星数据在农业土地信息提取中可以发挥重要作用,而由于农业土地具有显著的物候差异性,时间序列数据的应用能够有效提高作物的提取精度。不同作物因为物候期的差异,对时间序列的需求有明显的差异性。为了确定时间序列数据对冬小麦提取精度的影响,利用归一化植被指数(NDVI),增强植被指数(EVI)作为构造时序影像的特征,将覆盖整个冬小麦生育期的所有Sentinel-2A/B数据,以3个不同的时间间隔(10,15和30 d)的均值合成,使用随机森林分类器对像素或地块进行分类,以对比不同时间间隔合成的时序影像在面向不同元素的冬小麦提取时的应用效果和提取效率。结果表明:①当合成Sentinel-2A/B影像的时间间隔缩短时,2种分类方法的冬小麦的提取精度都会更高,但面向地块的分类方法能达到更好的提取效果;②10 d合成的面向地块的分类方法获得了最高的OA(96.28%)和Kappa系数(91.71%);③时间序列的合成方法对提取效率影响不大,影响它的主要原因是面向不同元素的分类方法。
Satellite data plays an important role in agricultural land information extraction.However,due to the significant phenological difference between agricultural lands,the application of time series data can effectively improve the accuracy of crop extraction.Different crops have different demands on time series because of their different phenological periods.In order to determine the influence of time series data on the accuracy of winter wheat extraction,NDVI and EVI are used as features to construct time series images,and all Sentinel-2A/B data covering the whole growth period of winter wheat are synthesized by the means of three different time intervals(10,15 and 30 d).Then,random forest classifier is used to classify the pixels or plots to compare the application effects and extraction efficiency of sequential images synthesized at different time intervals in winter wheat extraction for different elements.The results show that:①When the time interval of Sentinel-2A/B image synthesis is shortened,the extraction accuracy of winter wheat is higher for both classification methods,but the parcel oriented classification method can achieve better extraction effect;②The highest OA(96.28%)and Kappa coefficient(91.71%)are obtained by the 10 d classification method;③The method of time series synthesis has little effect on extraction efficiency,and the main reason is the classification method for different elements.
作者
石娴
明艳芳
刘春秀
瞿渝
隋淞蔓
SHI Xian;MING Yanfang;LIU Chunxiu;QU Yu;SUI Songman(School of Geomatics and Spatial Information,Shandong University of Science and Technology,Qingdao 266590,China)
出处
《无线电工程》
北大核心
2021年第12期1567-1576,共10页
Radio Engineering
基金
国家重点研发计划资助项目(2019YFE0126700)。
