Study On Physiological And Biochemical Parameters Of Maize Based On Spectral Data Of Ground And Low Altitude UAV

The physiological and biochemical parameters of crops are not only a visual expression of their nutrition,quality and health status but also an important reference for crop field management.With the continuous development of precision science and technology,the acquisition of crop physiological and biochemical parameters based on remote sensing platform has gradually replaced the traditional monitoring of physical and chemical attributes based on experimental methods,so as to become an important technique of the information management and ecological environment monitoring of farmland.In this thesis,corn with a wide planting range in Guanzhong area is taken as the research object,and the ground remote sensing and low altitude UAV remote sensing are used as the main technologies.Based on the ground hyperspectral,UAV multispectral,and UAV hyperspectral data of corn in the key growth period,as well as through spectral data processing and image information extraction,the characteristic bands and vegetation indexes sensitive to corn growth are selected,and the physiological and biochemical parameters of corn are estimated.As a result,it can provide noy only a reliable data basis for the refined management of corn but also a reference for the estimation of corn yield as well as the formulation of agricultural products import and export trade policy.The main contents and conclusions are as follows:(1)The spectral estimation models of Chla,Chlb,Cars,and Tot of maize in the whole growth period were constructed,so as to realize the accurate evaluation of the photosynthetic capacity of maize.Among the four photosynthetic pigment UR models based on the general sensitive parameters of Chla,Chlb,Cars,and Tot,the polynomial model has the best estimation effect,and the Cars polynomial model based on PSSR_c has the highest accuracy(R~2_c=0.62,RMSE_c=0.07;R~2_v=0.64,RMSE_v=0.07);In the MSR models of Chla,Chlb,Cars,and To T,the modeling effect is generally good,and all of them pass the 0.01 significance level test(R~2_c>0.72),but the correction effect is relatively poor(R~2_v>0.57);The model R_c~2 in the RF model of Chla,Chlb,Cars,and Tot is more than 0.9,and the test R_v~2 is more than 0.7.Compared with the UR and MSR models,the modeling and test accuracy of RF are significantly improved(p_c=0.001,p_v=0.02),which can realize the refinement and high-precision research of photosynthetic pigments of maize.(2)The estimation effects of UHD spectrum of UAV and of SVC spectrum ground on the SPAD of maize were compared,and a high-precision prediction model for the SPAD of maize canopy as well as its spatial distribution was constructed.For the SPAD estimation model of corn based on the sensitive bands extracted from the original spectrum and FD spectrum of the ground and low altitude UAV remote sensing platforms,the FD spectrum of the same remote sensing platform has a better fitting effect on SPAD than the original spectrum.Among the SPAD models of two single vegetation indexes based on remote sensing data,SVC spectral model and UHD-FD spectral model have better fitting effects.Among the MSR models of multiple vegetation index of two remote sensing data,the SVC spectral model(R_c~2=0.85,R_v~2=0.61,v=18)has the best fitting effect with respect to the ground spectrum;The UHD-FD spectral model(R_c~2=0.79,R_v~2=0.59,v=11)has a good spectral fitting effect for UAV.Although both the remote sensing platform data can realize the high-precision simulation of SPAD for corn,the number of variables in the UAV model is much less than that of the another.Moreover,the MSR model based on UHD-FD spectrum is also effective in predicting the spatial distribution of SPAD for corn in the whole growth period.Therefore,taking the accuracy of model,the monitoring range,and the amount of data calculation into consideration,UHD-FD spectrum can be used as a substitute for ground spectrum to achieve high-precision prediction of SPAD for maize canopy and its spatial distribution.(3)Data mining of ground spectra and UAV images is carried out from the perspectives of spectral transformation(PCA,FD,and CR),image processing,and texture extraction,etc.,so as to construct a robust multi-source remote sensing collaborative estimation model of Anth for corn canopy.Among the different transform ground hyperspectral models,the fitting effect of spectral parameters of FD transformation on Anth is better than that of PCA and CR transformations.Among the ground remote sensing models with different spectral transformations,RF method is better than MSR,PLS,and BPNN in modeling and testing Anth(FD-RF:R_c~2=0.91,R_v~2=0.51).At the same time,the collaborative estimation model(R_c~2=0.93,R_v~2=0.76)combining the best ground transform spectrum(FD spectrum)and UAV image significantly improve the fitting accuracy of Anth compared with a single remote sensing data source(UAV or ground remote sensing).This is conducive to the accurate monitoring Anth of corn.(4)The potential about UHD spectrum of UAV and SVC spectrum of ground,as well as the original spectrum and FD spectrum of two platforms in estimating the LAI of corn are analyzed,such that a high-precision prediction model for LAI of corn and its spatial distribution is constructed.Among the UR models of spectral sensitive bands for SVC,UHD,SVC-FD,and UHD-FD,the 674 nm reflectance of UHD spectrum has the best estimation effect on LAI(R_c~2=0.47,R_v~2=0.46),but the accuracy of the model is low.Among the UR models based on vegetation index,the estimation accuracy of SIPI power function model for UHD spectral is the highest,but the prediction of LAI spatial distribution characteristics is not accurate enough.Among the MSR models based on vegetation index,the two remote sensing platform models with better LAI estimation effects are the MSR model of SVC spectral(R_c~2=0.69,R_v~2=0.64,v=16)and the MSR model of UHD spectral(R_c~2=0.66,R_v~2=0.65,v=8).Although the fitting accuracy of ground and UAV data for LAI is similar,the number of variables in UAV model is less and the model is simpler than that of ground.At the same time,the MSR model based on UHD-FD spectrum can also have a better prediction effect on the spatial distribution of LAI in the whole growth period of maize.Therefore,by considering the accuracy of model,the amount of data,and the monitoring range,UHD spectrum can replace the ground spectrum for high-precision estimation of LAI and its spatial distribution in maize canopy.(5)A high-precision classification model of healthy and red leaves of maize was constructed.Through the high-precision estimation of anthocyanin content in leaves,the indirect estimation of the degree of redness of red leaves and the risk of redness of healthy leaves of maize was realized,which provided a new idea for the remote sensing research of plant diseases and pests.The results showed that the maximum value of spectral difference between red and healthy leaves appeared near 700 nm,and the maximum value of spectral difference index DI can exceed 3.The accuracy of LDA and SVM classification models based on specific vegetation index VI_c is significantly higher than that of narrow band vegetation index VI_s and characteristic band Rλ,so as to realize the complete recognition of red and healthy leaves for corn(CA_c=CA_v=100%).Among the estimation models of red leaf for Anth,the MLR model based on R_λ+VI_s+VI_c has the best estimation effect(R_c~2=0.85,R_v~2=0.74).Among the estimation model of healthy leaves for Anth,the SVM model based on R_λ+VI_s+VI_c has the best estimation effect(R_c~2=0.68,R_v~2=0.66).Both can be used to perform high-precision estimation of Anth content in two kinds of leaves.The above research proved the great potential of ground hyperspectral remote sensing system,UAV visible light system,and UAV hyperspectral system in obtaining physiological and biochemical parameters of maize,and also provided technical reference for obtaining refined crop growth information.