Summer Corn Growth Monitoring And Yield Estimation By UAV-based Multispectral Remote Sensing

To facilitate rapid and accurate monitoring of crop growth information and estimation of crop yield in agricultural production,this study takes summer maize,a major crop in China,as the research object,and uses UAV multispectral images to obtain vegetation indices related to summer maize growth,and uses various vegetation indices as independent variables and the relative chlorophyll content,leaf area index,above-ground biomass,plant water content and yield of summer maize at four reproductive stages,namely,nodulation,tasseling,lactation and maturity,as independent variables.Biomass,plant water content growth index and yield as the dependent variables,the best-fit models for growth monitoring and yield estimation of summer maize at each fertility stage were developed and the related models were tested with independent data,and the results are as follows.（1）Among the models for monitoring the relative chlorophyll content of summer maize in the four reproductive stages,the best results were based on the monitoring models constructed by NDVI,GRVI,GRVI,and CCCI with R~2 of 0.612,0.577,0.602,and 0.419,respectively,and the SPAD monitoring was better in the early growth stage and not suitable for SPAD monitoring in the later stage.（2）Among the four fertility summer maize leaf area index monitoring models,the best results were based on the monitoring models constructed for RVI,RVI,NGRDI,and NDVI,with R~2 of 0.641,0.606,0.592,and 0.533,respectively.it was concluded that the tasseling and lactation stages were suitable for LAI monitoring.（3）Among the four fertility stages of summer maize aboveground biomass monitoring models,the best results were based on the monitoring models constructed by NGRDI,RVI,NDVI,and RVI with R~2 of 0.561,0.613,0.616,and 0.583,respectively,which were suitable for aboveground biomass monitoring in the middle of summer maize growth.（4）Among the four fertility stages of summer maize,the best monitoring models for plant water content monitoring were those constructed based on RGRI,RGRI,NGRDI,and NDVI,with R~2 of 0.571,0.633,0.592,and 0.624,respectively,for all fertility stages.（5）The coefficient of variation method was used to construct comprehensive growth monitoring indicators as the dependent variable,and three methods,namely partial least squares regression,random forest and BP neural network,were adopted to construct comprehensive growth monitoring models for the four fertility stages of summer maize.The best model was the random forest method with R~2 of 0.746 at the tassel stage.（6）In the summer maize yield estimation model,the correlation between various vegetation indices and yield was poor in the early stage of growth,making it difficult to estimate yield in the early stage of growth;while in the middle and late stage of growth,the regression model constructed using vegetation indices could effectively estimate the yield of summer maize,and the R~2 of the monitoring model constructed based on RVI,DVI,and SAVI was 0.457 in the milk ripening stage;in the maturity stage,the NDVI,RVI,DVI,and SAVI multivariate linear fitting model R~2 reached 0.673.The models constructed at each fertility stage were more suitable for monitoring chlorophyll content,RVI and NGRDI for monitoring LAI,RVI for monitoring aboveground biomass,and RGRI for monitoring plant water content.In monitoring chlorophyll content and LAI,the early stage was better and the later stage was less satisfactory;the middle stage of maize growth was suitable for monitoring aboveground biomass;the monitoring of plant water content was not affected by the reproductive stage.This study can provide a useful reference for growth monitoring and yield estimation of summer maize.