Study On Remote Sensing Inversion Of LAI And Chlorophyll Content In Maize Based On Hyperspectral Data

Nowadays,international modern agriculture pays more and more attention to the accurate diagnosis and management of crop production.It pays more attention to the efficient and accurate management while having high yield.Remote sensing observation plays a key role in the development of precision agriculture by providing accurate crop biophysical and biochemical variables,such as leaf area index(LAI)and leaf chlorophyll content(Cab).Therefore,quantitative estimation of the content of physical and chemical parameters of crops is of great significance for monitoring crop growth,ecological environment change and global climate change.In recent years,hyperspectral remote sensing technology has been widely used in crop production and management,and radiation transfer model has also become a widely used physical inversion method.However,it remains to be studied whether to consider the influence of male tassel on the radiation transfer characteristics of the canopy when using radiation transfer model for crops with spikes.The vertical division of chlorophyll content and leaf area index within the canopy is still to be studied Whether the spatiotemporal dynamic change of cloth has a certain impact on remote sensing monitoring needs further study,and whether there are other better inversion methods that can be used in UAV hyperspectral remote sensing for the inversion of physical parameters is still uncertain.In this paper,maize is taken as the main research object.Through the analysis and research of four kinds of physicochemical parameter inversion methods(vegetation index regression method,nonparametric regression method,physical method and hybrid method)which are mainly used in the research at home and abroad at present,the influence of tassel stage on the radiation transmission characteristics of the canopy based on ASD near ground multi angle canopy hyperspectral data is studied The main conclusions are as follows:(1)The results showed that tassel had an effect on the radiative transfer characteristics of canopy reflectance at heading stage.a)the global sensitivity analysis of radiative transfer model at heading stage showed that leaf area index and chlorophyll content were more than80%sensitive to the model,which were the main factors.According to the analysis of LAI and Cab measured in the whole heading stage of maize,the difference between LAI and Cab is small,the change range of LAI is 1.23-1.53 m~2/m~2,the overall shape of all points tends to straight line,and the change range of Cab between sample points is 1.25-6.7?g/cm~2;b)compared the simulated and measured values of canopy reflectance at different time of heading stage,the measured values were higher than the simulated values at the beginning of heading stage,and the simulated values were higher than the measured values gradually with the development of growing stage.However,in the visible and near-infrared bands,the measured values are higher than the simulated values in the whole heading stage;c)the analysis of the two-way reflectance of the canopy layers with different tassel gradients shows that in the visible and near-infrared bands,the observed reflectance in the two scattering directions and in the vertical observation direction,the observed reflectance generally shows no tassel>1/2 tassel>whole tassel;d)PROSAIL model simulation values and LAI and Cab showed a significant negative correlation in the whole band,and the correlation between LAI and Cab was consistent with the measured and simulated values of no panicle;e)there was a significant difference in the change of the total fresh weight of the male panicle during the whole heading period,while the change of the total dry weight was not significant.The water content of male tassel decreased from about 80%at the beginning of heading to about 10%at the end of heading.So in conclusion,when using radiation transfer model to simulate canopy reflectance at heading stage,the model should be revised.It is considered that the correlation between water content of male tassel and equivalent water thickness of leaves as input parameters of PROSAIL model can be taken as a new input parameter to participate in model simulation operation;(2)Based on the multi angle canopy hyperspectral data set near the ground,the potential of four inversion methods for Lai and cab was analyzed by using ARTMO software.The results show that:a)the hybrid method of MLRAs and RTMs(MLRA_RTMs)has the highest inversion accuracy and the best performance;the vegetation index inversion method takes the second place,but the fastest operation speed;the physical model inversion based on LUT_RTM The accuracy is the lowest;b)vegetation index inversion method has better inversion results for LAI and Cab in backscattering direction(-50?),while the inversion results of Canopy Chlorophyll content(CCC)based on LUT_RTM inversion method are the best under multi angle observation;c)among the three nonparametric regression methods,kernel ridge regression(KRR)and Gaussian process regression(GPR)have better estimation results for LAI,Cab and CCC,and three agricultural regression methods have better estimation results The inversion accuracy of the parameters is between 0.68 and 0.83.Especially based on single angle observation,KRR and GPR in the backscatter direction(-50?)have the best inversion results for LAI and Cab respectively,while GPR in the forward scattering direction(+50?)has the best inversion results for CCC;d)hybrid method based on MLRA_RTM has the best inversion results in four inversion methods,and multi angle(0?,-50?,+50?)and Both KRR_RTM and GPR_RTM observation data improve the estimation results of three parameters,but both reduce the operation speed;(3)The quality of hyperspectral data set of UAV is verified and analyzed by many methods,and it is concluded that the quality of the data set is reliable and can be used as remote sensing inversion of physical and chemical parameters of maize.Then,four inversion methods are used to inversion LAI,Cab and CCC,and the research results are as follows:a)the inversion accuracy of three agricultural parameters of maize based on MLRAs is the best,and the hybrid method of MLRA_RTMs is the second The inversion accuracy based on LUT_RTM and SI is the same and the lowest;b)the inversion method of vegetation index is the fastest,(MLRA_RTMs)hybrid method is the second,MLRAs and LUT_RTM inversion method has the same and slowest operation speed;c)vegetation index model has the highest certainty,and GNDVI has the best correlation with the three parameters,with the determination coefficients of 0.769,0.818 and 0.888 respectively;polynomial model has the best estimation effect on LAI,power model has the best inversion effect on Cab and CCC,and the fastest operation speed;d)LUT_RTM is used in inversion The relative error can be significantly reduced by adding some noise and multiple solutions to the inversion of physical and chemical parameters of maize,and when the data is normalized,the relative error can be significantly reduced by using the method of M_estimate L1_estimate can obtain the best Cab estimation results(nrmse is 18.33%at 7%multiple solutions and 20%noise);however,data normalization is not conducive to inversion of LAI,and LSE produces the best inversion results when using non normalized data(NRMSE is 14.12%at 2%multiple solutions and13%noise).When the same method is used for CCC inversion,the inversion result is similar to LAI;e)MLRAs has the best estimation result for Cab,MLRA_RTMs has the same estimation result for LAI and CCC,GPR has the better estimation result for Cab and CCC,and KRR has the better estimation result for LAI;(4)By using the model calibration and validation data set of the test area,the established optimal model of remote sensing inversion of physical and chemical parameters of maize based on near ground multi angle observation data set and UAV hyperspectral image data set is validated and evaluated.The results show that KRR_PROSAIL and GPR_PROSAIL are the best remote sensing inversion models of maize leaf area index,chlorophyll content and Canopy Chlorophyll content respectively,while KRR and GPR in machine learning regression algorithm based on UAV hyperspectral observation are maize leaf area index,chlorophyll content and canopy green respectively The best remote sensing inversion model of element content.This chapter shows that the inversion ability of MLRAs is the best under the two platforms.This inversion method has certain application potential when applied to the hyperspectral image of UAV for the inversion of physical and chemical parameters of maize.