摘要
针对东北粳稻叶绿素含量无人机高光谱反演中红边位置特征不明确的问题,基于2019~2020年沈阳农业大学水稻试验基地采集的高光谱数据和地面水稻样本叶绿素数据,开展水稻叶绿素含量红边光谱响应特性分析及反演建模研究。首先,利用线性外推(LE)、线性四点插值(LI)、最大一阶导数(MFD)、多项式拟合(PF)、拉格朗日插值(LAG)和倒高斯拟合(IG)6种方法确定水稻冠层高光谱红边位置,对比不同方法提取结果的分布规律,分析叶绿素含量的红边位置响应特性。然后,基于不同红边位置提取结果,利用极限学习机(ELM)、反向传播神经网络(BPNN)两种机器学习算法以及5种统计回归算法构建水稻叶绿素含量反演模型。结果表明:LE方法适合东北粳稻冠层光谱的红边位置的计算,提取的红边位置变幅大,对叶绿素含量变化较为敏感;以LE提取的红边位置为输入特征,构建的线性外推-极限学习机(LE-ELM)和线性外推-对数曲线(LE-LNX)模型,叶绿素含量反演精度较高,两个模型决定系数、均方根误差、平均绝对误差分别为0.781,8.375,9.828和0.763,9.249,11.253,研究结果可为水稻冠层叶绿素含量高效监测提供理论支撑。
The red edge spectral response characteristics analyzing and inversing of rice(Oryza sativa L.ssp.japonica)chlorophyll content were proposed in the paper,aiming to solve the problem of unclear red edge position in UAV hyperspectral.Hyperspectral and the related chlorophyll contents of rice samples were collected from the Rice Research Base of Shenyang Agricultural University in 2019 and 2020.First,the red edge position in the rice canopy hyperspectral was determined by using linear extrapolation(LE),linear four-point interpolation(LI),maximum first derivative(MFD),polynomial fitting(PF),Lagrange interpolation(LAG)and inverse Gaussian fitting(IG).The response characteristics of the red edge of the chlorophyll content were analyzed,based on comparing of distribution characteristics of the extraction results of the different methods.Then,based on the extraction results of different red edge positions,two nonlinear machine learning algorithms,extreme learning machine(ELM),back propagation neural network(BPNN),and five statistical regression algorithms were used to construct a rice chlorophyll content inversion model.The results showed that the LE method is suitable for the calculation of the red edge position of the northeast Japonica rice canopy spectrum.The extracted red edge position had a large amplitude and was more sensitive to the change of chlorophyll contents.The red edge position extracted by LE was the input feature,which is a linear extrapolation constructed.The accuracy of chlorophyll content inversion model of extreme learning machine(LE-ELM)and linear extrapolation-logarithmic curve(LE-LNX)are topgallant,with the determination coefficient,root mean square error,and average absolute error of 0.781,8.375,9.828 and 0.763,9.249,11.253,respectively,which can provide theoretical support for the efficient monitoring of rice canopy chlorophyll content.
作者
曹英丽
江凯伦
刘亚帝
于正鑫
肖文
于丰华
CAO Ying-li;JIANG Kai-lun;LIU Ya-di;YU Zheng-xin;XIAO Wen;YU Feng-hua(School of Information and Electrical Engineering/Liaoning Agricultural Information Engineering Technology Center,Shenyang Agricultural University,Shenyang 110161,China)
出处
《沈阳农业大学学报》
CAS
CSCD
北大核心
2021年第6期718-728,共11页
Journal of Shenyang Agricultural University
基金
国家重点研发计划项目(2017YFD0300706)
辽宁省教育厅课题重点项目(LSNZD201605)。
关键词
无人机遥感
高光谱
红边位置
水稻叶绿素
叶绿素反演
UAV remote sensing
hyperspectral
red edge position
rice chlorophyll
chlorophyll retrieval
