| Winter wheat is an important food crop second only to rice and maize.As important agronomic parameters reflecting the growth of winter wheat,biomass and leaf area index can reflect the crop growth status and play a guiding role in yield monitoring and estimation.Taking winter wheat as the research object,this paper analyzes the canopy spectral characteristics of Winter Wheat under different nitrogen conditions by using hyperspectral remote sensing data,mathematical statistics and hyperspectral data transformation,and establishes the estimation models of winter wheat biomass and leaf area index based on spectral parameters,first derivative spectrum and wavelet energy coefficient by using machine learning algorithm.The main research contents and conclusions of this paper are as follows:(1)The changes of biomass and leaf area index of winter wheat at different growth stages and the spectral response characteristics under different nitrogen conditions were studied.The results showed that the spectral reflectance of winter wheat canopy decreased first and then increased in the visible band,and increased first and then decreased in the near-infrared band.With the increase of biomass and leaf area index,the canopy spectral reflectance of winter wheat decreases with the increase of biomass and leaf area index in visible wave band,and increases with the increase of biomass and leaf area index in near-infrared band.In the four growth stages,the canopy spectral reflectance in visible band decreased with the increase of nitrogen level,and the reflectance in near-infrared band increased with the increase of nitrogen level.(2)Based on the hyperspectral data and biomass data of winter wheat canopy,a winter wheat biomass estimation model based on spectral parameters,first derivative spectrum and wavelet energy coefficient was established.The correlation of spectral parameters,first derivative spectra and wavelet energy coefficients with winter wheat biomass at different growth stages was analyzed.The results showed that the ratio of red edge area to blue edge area had the strongest correlation with biomass at jointing stage and booting stage,and the ratio of red edge area to vegetation index and red edge amplitude had the strongest correlation with biomass at flowering stage and filling stage;The sensitive bands of the first derivative spectrum to winter wheat biomass are concentrated in the visible near infrared range,and the bands with good correlation in each growth period are 956 nm,768 nm,1133 nm and 749 nm respectively;The correlation between wavelet energy coefficient and biomass increases first and then decreases with the increase of decomposition scale.The optimal band scale is mainly concentrated in 3,4,5 and 6;In jointing stage and filling stage,the accuracy of winter wheat biomass estimation of recursive feature elimination Gaussian process regression model is the best,and the verification R~2 is 0.87 and 0.84 respectively.At booting stage,the estimation accuracy of winter wheat biomass based on the first derivative spectrum support vector machine model is the best,and the verification R~2 is 0.75.At the flowering stage,the accuracy of winter wheat biomass estimation based on recursive feature elimination random forest model is the best,and the verification R~2 is 0.87.(3)Construction of estimation model of winter wheat leaf area index based on spectral parameters,first derivative spectrum and wavelet energy coefficient.The correlation between spectral parameters,first derivative spectra and wavelet energy coefficients in different growth stages and winter wheat leaf area index was analyzed.The results showed that in jointing stage and flowering stage,the spectral parameter with the strongest correlation with leaf area index was water index,and the red edge ratio vegetation index was in booting stage and filling stage;The sensitive band of the first derivative spectrum to the leaf area index of winter wheat is mainly concentrated in the visible near infrared band,and the bands with good correlation in each growth period are 955nm,756nm,1127nm and 742nm respectively;The correlation between wavelet energy coefficient and leaf area index increases first and then decreases with the increase of decomposition scale.The scale of the selected optimal band is mainly concentrated in 1,2,3 and 5.In jointing stage,booting stage and filling stage,the estimation accuracy of winter wheat leaf area index of recursive feature elimination random forest model is the best,and the verification R~2is 0.85,0.82 and 0.88respectively;At the flowering stage,the recursive feature elimination Gaussian process regression model is the best,and the verification R~2 is 0.90.This paper has 30 figures,14 tables and 89 references. |
