基于Sentinel-2遥感影像的玉米秸秆覆盖度估算研究  

Estimation of corn residue coverage based on Sentinel-2 remote sensing images

作　　者：韩颖[1] 张龙龙 柳阳 冯仰强 HAN Ying;ZHANG Long-long;LIU Yang;FENG Yang-qiang(University of Science and Technology Liaoning,Anshan 114051,China;Lanzhou Aerospace Hongtu Information Technology Co.,Ltd.,Lanzhou 730010,China)

机构地区：[1]辽宁科技大学,辽宁鞍山114051 [2]兰州航天宏图信息技术有限公司,甘肃兰州730010

出　　处：《环境生态学》2024年第11期43-51,共9页Environmental Ecology

基　　金：国家重点研发计划项目子课题“耕地利用与保护模式监测研究”(2021YFD1500103);中国科学院战略性先导科技专项课题(XDA28080500);辽宁省教育厅高校基本科研项目(LJ242410146047)资助。

摘　　要：作物秸秆作为农业生产活动中的重要物质,在减少土壤侵蚀、提高土壤保熵及增加土壤团聚体等方面发挥着重要的作用,因此通过遥感技术对大范围作物秸秆覆盖度进行监测具有重要的意义。本研究以样线法和图像分割法2种方法对玉米秸秆覆盖度(Crop Residue Cover,CRC)进行野外采样测量,以Sentinel-2影像为基础,利用遥感光谱指数和纹理特征作为特征变量,选择偏最小二乘法(Partial Least Squares Regression,PLSR)建立大范围CRC估算模型。结果表明:1)从遥感影像上提取的遥感光谱指数和纹理特征均与CRC有较好的相关性,其中,简单耕作指数(Simple Tillage Index,STI)与CRC相关性优于其他遥感光谱指数,样线法和图像分割法的决定系数(Coefficient of Determination,R^(2))分别为0.844和0.848;纹理特征与CRC的相关程度低于遥感光谱指数,其中B8A Mean纹理特征与玉米CRC相关性优于其他纹理特征,样线法和图像分割法的R 2分别为0.505和0.507。2)以遥感光谱指数为主、纹理特征为辅的组合方法构建的PLSR模型用于玉米CRC估算精度高于采用单一遥感光谱指数或纹理特征建模的CRC估算精度;其中样线法建模结果相关系数R 2达到0.896,图像分割法构建的模型相关系数R 2为0.892,因此,利用分辨率较高的Sentinel-2影像构建的遥感光谱指数和纹理特征建立PLUS模型在CRC估算方面精度较好,具有良好的应用前景。As an important material in agricultural production activities,crop residue plays a key role in reducing soil erosion,improving soil entropy retention,and increasing soil aggregates and organic matter;Hence,how to accurately estimate its coverage by remote sensing is of great importance.This study was based on the Sentinel-2 image,selected the sample rope method and the image segmentation method to measure the Crop Residue Cover(CRC)on the ground,used remote sensing spectral indexes and texture feature as characteristic variables,evaluated the estimation ability of Sentinel-2 images in CRC through comparative analysis.The results show that:1)The correlation between CRC and Simple Tillage Index(STI)is better than other remote sensing spectral indexes.The coefficient of determination(R 2)of line transect method and image segmentation method were 0.844 and 0.848,respectively.The correlation between texture features and CRC was lower than that of remote sensing spectral index,and the correlation between B8A Mean texture features and CRC was better than other texture features.The R 2 of sample rope method and image segmentation method were 0.505 and 0.507.2)The accuracy of PLSR model constructed with remote sensing spectral indexes as the main method and texture feature as the secondary method for large area CRC estimation is higher than that of CRC model established with single remote sensing spectral index or texture feature.The correlation coefficient R 2 of line transect modeling results is 0.896,and the correlation coefficient R 2 of image segmentation model is 0.892.Therefore,the PLSR model established by using the combination of remote sensing spectral indexes and texture features extracted from Sentinel-2 remote sensing images has good accuracy in CRC estimation,and has a good application prospect.

关 键 词：秸秆覆盖度 遥感光谱指数 纹理特征 样线法 图像分割法 偏最小二乘法 

分 类 号：P237[天文地球—摄影测量与遥感]