基于Sentinel-2影像的河套灌区作物种植结构提取

作物识别是提取作物种植结构的基础,利用遥感技术对作物进行监测识别,对优化生产布局、调整农业生产模式有着重要意义。文中选取河套灌区杭锦后旗为研究区域,基于2019年覆盖生长周期的Sentinel-2号卫星影像数据,构建NDVI时间序列数据集,利用Savitzky-Golay(S-G)滤波对NDVI时间序列数据集进行平滑,分析不同作物不同发育期的光谱曲线特征,计算各主要作物识别关键期的光谱阈值,构建基于决策树分层分类的农作物种植面积提取模型,并用验证样本对分类结果进行精度验证。结果表明:利用整个生育期内的NDVI最大合成影像确定植被地表覆盖,NDVI曲线变化区别林地与耕地,逐层提取地物,简便易行;采用S-G滤波重构高质量的NDVI时间序列曲线,研究证明重构后曲线更加平滑符合作物生长趋势;基于Sentinel-2号遥感数据和整个生育期NDVI时序数据,构建分层分类决策树模型,作物分类总体精度达92.1%,Kapppa系数精度达0.857。本研究采用的方法满足遥感观测应用化需求,也为县级区域农作物分类提供重要参考价值。

Extraction of crop planting structure in Hetao irrigated area based on Sentinel-2

Crop identification is the basis of crop planting structure extraction.Using remote sensing technology to monitor and identify crops is of great significance to optimize production layout and adjust agricultural production mode.In this study,Hangjin houqi was selected as the study area,based on images from the Sentinel-2 satellite,which will cover the growth cycle in 2019,the NDVI time series data set was built,the Savitzky-Golay(S-G)filter was used to smooth the NDVI time series data set,the characteristics of the spectral curve of different crops in different developmental stages were analyzed,spectrums of major crops were computed to identify the origins of threshold,to build crop planting area extraction model on the basis of the decision tree classification model,and the accuracy of the classification result validation sample was verified.The results showed that:The maximum NDVI composite image of the whole growth period could be used to determine the vegetation surface cover.Using NDVI curve change to distinguish between forest land and cultivated land is simple and easy to extract ground objects layer by layer.The high quality NDVI time series curve was reconstructed by s-G filtering,and the results showed that the reconstructed curve is smoother and in line with the trend of crop growth.A hierarchical classification decision tree model was constructed based on sentinel-2 remote sensing data and the NDVI timing sequence data during the whole growth period.The overall accuracy of crop classification is 92.1%,and the Kapppa coefficient accuracy is 0.857.The methods adopted in this study meet the needs of remote sensing observation application and also provide important reference value for crop classification in county-level regions.