摘要
高光谱遥感技术为水稻叶片叶绿素含量的高通量、无损、准确监测提供了有效途径,然而高光谱数据的降维或特征光谱参数的选择是叶绿素含量有效反演的关键环节。利用2017年辽宁省盘锦市大洼水稻氮高效品种筛选试验基地的水稻叶片叶绿素含量与叶片高光谱数据,探讨了高光谱数据的降维方法与叶绿素含量的反演建模。首先应用最优子集选择算法(best subset selection)对工程常用的水稻叶绿素反演特征光谱指数进行优选,筛选出最优组合,作为叶绿素多元回归模型的输入特征;同时应用没有在光谱领域得到有效应用的基函数展开算法,利用Gram-Schmidt正交变换寻找叶片高光谱数据的基函数空间,再将高光谱数据投影到基函数空间从而实现降维,最后利用降维后的数据进行多元回归建模,反演叶绿素。结果表明:最优子集选择算法优选出的mNDVI(445,705,750)、NDVI(705,750)、PSRI(500,680,750)、RD(505,705)、RI1dB(720,735)、MCARI(550,670,700)、PPR(450,550)共7个特征指数组合,回归模型反演精度最高,决定性系数R2为0.844,均方根误差RMSE为0.926;基于基函数展开算法对400～1000nm波段范围601维高光谱数据降至13维,叶绿素反演回归模型的决定性系数R2达到0.861,均方根误差RMSE为0.906。说明基于基函数展开的高光谱降维与叶绿素含量估测方法效果较好,可为水稻叶绿素含量估测与长势诊断提供技术支持。
Hyperspectral remote sensing technology provides an effective way for high-throughput,non-destructive and accurate monitoring of chlorophyll content in rice leaves.However,the reduction of hyperspectral data or the selection of characteristic spectral parameters is the key link for effective retrieval of chlorophyll content.In this study,the rice leaf chlorophyll content and leaf hyperspectral data from Dawa County,Panjin,Liaoning Province in 2017 were used to explore the dimensionality reduction methods and chlorophyll content inversion modeling of hyperspectral data.First,the best subset selection algorithm was used to optimize the spectral index of rice chlorophyll inversion,which was commonly used in engineering.The optimal combination was selected as the input feature of the chlorophyll multivariate regression model.Meanwhile,the Gram-Schmidt orthogonal transformation was applied.To find the basis function space of the hyperspectral data of the blade,the hyperspectral data were projected to the basis function space to achieve dimensionality reduction.Finally,the reduced-dimensional data were used to perform multiple regression modeling to invert the chlorophyll.The results showed that seven characteristic index combinations of mNDVI(445,705,750),NDVI(705,750),PSRI(500,680,750),RD(505,705),and RI1dB(720,735),MCARI(550,670,700),PPR(450,550)fitted to the regression model with the highest inversion accuracy,the decisive coefficient R2 was 0.844,the root mean square error RMSE was 0.926.Based on the basis function Expanding the algorithm,the 601-dimensional hyperspectral data in the 400-1000nm band range was reduced to 13-dimensional,and the deterministic coefficient R2 of the chlorophyll inversion regression model reached 0.861,and the root mean square error RMSE was 0.906.It showed that the hyperspectral dimension reduction and chlorophyll content estimation method based on the basis function expansion was effective.So it can provide technical support for rice chlorophyll content estimation and growth diagnosis.
作者
曹英丽
邹焕成
郑伟
江凯伦
于丰华
CAO Ying-li;ZOU Huan-cheng;ZHENG Wei;JIANG Kai-lun;YU Feng-hua(Liaoning Agricultural Information Engineering Technology Center/College of Information and Electrical Engineering,Shenyang Agricultural University,Shenyang 110161,China)
出处
《沈阳农业大学学报》
CAS
CSCD
北大核心
2019年第1期101-107,共7页
Journal of Shenyang Agricultural University
基金
国家重点研发项目(2017YFD0300706
2016YFD0200700)
辽宁省教育厅课题重点项目(LSNZD201605)
关键词
叶绿素
高光谱降维
基函数展开
最优子集选择
特征光谱指数
水稻叶片
chlorophyll
hyperspectral dimension reduction
basis function expansion
optimal subset selection
characteristic spectral index
rice leaf
