Study On The Estimation Models Of Winter Wheat Growth Parameters And Yield Based On Hyperspectral Remote Sensing

Winter wheat is one of the three major food crops in the world.Chlorophyll content,leaf area index and biomass are the key indicators for crop growth monitoring and yield prediction.Therefore,rapid,nondestructive and accurate monitoring of crop growth parameters and yield is of great significance for mastering crop growth status in time and ensuring national food security.The traditional measurement methods of crop growth parameters and yield are not only time-consuming and laborious,but also destructive in the field,and have low popularization and application value.In recent years,the wide application of hyperspectral remote sensing technology in crop growth monitoring has provided an effective way for the development of modern precision agriculture.This paper took winter wheat as the research object,and took the winter wheat comprehensive experimental field in Shizhuang village,Ningguo Town,Shanyang District,Jiaozuo City,Henan Province as the research area.The canopy hyperspectral reflectance data of winter wheat at different growth stages were obtained based on ASD ground object hyperspectral spectrometer.The original hyperspectral data were processed by fractional differential and continuous wavelet transform technology respectively.The fractional differential spectrum and wavelet energy coefficient with strong correlation with chlorophyll content,leaf area index and biomass were screened by correlation analysis.Then,support vector machine,optimal subset regression and BP neural network are used to construct the estimation models of winter wheat growth parameters,and the models accuracy were evaluated to screen the optimal estimation models.Finally,based on the estimation of growth parameters,spectral parameters and spectral parameters coupling growth parameters were used as the input characteristics of the models.The yield prediction models of single growth period and multi growth period of winter wheat were constructed based on elastic network algorithm and stacking technology,and the models accuracy were verified to screen the optimal yield prediction models.It provided theoretical reference and technical support for crop growth monitoring and yield prediction.The main conclusions of the paper are as follows:(1)The estimation models of winter wheat growth parameters was constructed by using fractional differential spectrum and wavelet energy coefficient combined with support vector machine,optimal subset regression and BP neural network.The results showed that,at the jointing stage,the best estimation accuracy could be obtained by using BP neural network combined with wavelet energy coefficient to construct chlorophyll content and biomass estimation models,and using support vector machine combined with wavelet energy coefficient to construct leaf area index estimation models,and the accuracy evaluation grades of the models were B,D and B respectively;At booting stage,the best estimation accuracy can be obtained by using support vector machine combined with wavelet energy coefficient to construct chlorophyll content and biomass estimation models,and using support vector machine combined with fractional differential spectrum to construct leaf area index estimation models,and the accuracy evaluation grade of the models is B;At flowering stage,the best estimation accuracy can be obtained by using BP neural network combined with fractional differential spectrum to construct chlorophyll content and biomass estimation models,and using optimal subset regression combined with wavelet energy coefficient to construct leaf area index estimation models,and the accuracy evaluation grades of the models are B,A and B respectively;at the filling stage,the best estimation accuracy can be obtained by using the optimal subset regression combined with wavelet energy coefficient to construct the estimation models of chlorophyll content and leaf area index,and using BP neural network combined with fractional differential spectrum to construct the biomass estimation models,and the accuracy evaluation grade of the models is B.On the whole,the filling period can be used as the best growth period for estimating the chlorophyll content and leaf area index of winter wheat,and the flowering period can be used as the best growth period for estimating the biomass of winter wheat.(2)Based on the estimation of growth parameters,the yield prediction models of single growth stage and multi growth stages of winter wheat were constructed by using elastic network algorithm and stacking technology respectively.The results showed that compared with only using spectral parameters,the accuracy and stability of the models constructed by coupling spectral parameters with growth potential parameters as input characteristics were effectively improved.The verification accuracy R~2 of yield prediction models at jointing stage,booting stage,flowering stage,filling stage and multi growth stage was increased by 5.12%,8.62%,8.79%,7.76%and 6.16%,and RMSE was reduced by 3.97%,7.08%,9.92%,10.02%and 13.09%respectively.Compared with only using single growth staged information,the yield prediction models constructed by stacking the prediction results of multiple growth stages had higher accuracy and stronger universality.Among them,the yield prediction models constructed by stacking of spectral parameters and growth parameters of four growth stages could obtain the best verification accuracy,R~2=0.80 and RMSE=607.04 kg/ha.The paper includes a total of 45 figures,28 tables,and 88 references.