基于ESTARFM模型的农作物类型识别方法应用

Application of crop type identification method based on ESTARFM model

【目的】利用时空融合技术生成高时空分辨率影像,为农作物类型识别研究提供一种思路和方法。【方法】以河套灌区部分区域(40°10′N~41°25′N、106°23′E~108°47′E)为研究区,基于MODIS和Landsat融合影像,利用增强型时空自适应反射率融合模型(ESTARFM),预测Landsat影像并建立NDVI时间序列影像数据集。结合地面样方数据将真实影像与预测影像数据集导入随机森林分类器,并对比分析研究区主要农作物类型。【结果】ESTARFM模型融合影像具有较清晰的空间表达能力,预测影像与真实影像波段表现出较好的相关性,其R值均能达到0.6以上。利用NDVI时间序列预测影像数据集农作物类型识别总体精度为93.03%,比真实影像精度高12.07%,Kappa系数为0.89。【结论】ESTARFM模型能够有效地解决农作物特定窗口期影像缺失困难,并能为农作物类型识别研究提供一种较好的应用方法。

【Objective】The spatiotemporal fusion technology was used to generate high spatiotemporal resolution images,which provides an idea and method for the research of crop type identification.【Methods】Taking part of the Hetao Irrigation Area(40°10′N-41°25′N,106°23′E-108°47′E)as the study area,based on MODIS and Landsat fusion images,using enhanced adaptive reflectance spatiotemporal fusion model(ESTARFM),to predict Landsat images and build NDVI time series image data set.Combining the ground sample data,the real image and predicted image data sets were imported into the random forest classifier,and the main crop types in the study area were compared and analyzed.【Results】The ESTARFM model fusion image had a clearer spatial expression ability,the predicted image and the real image had a better correlation between the bands,and the R value could reach more than 0.6.The overall accuracy of crop type identification using NDVI time series prediction image data set was 93.03%,which was 12.07% higher than the real image accuracy,and the Kappa coefficient was 0.89.【Conclusion】The ESTARFM model could effectively solve the difficulty of image missing in a specific window period of crops,and provide a better application method for crop type identification research.