Research On Wheat Growth Monitoring And Yield Estimation Based On Drone Image Data

With the continuous development of smart agriculture,remote sensing technology is increasingly used in agricultural production,mainly in crop growth monitoring,yield estimation and pest and disease control.Crop growth monitoring and yield estimation play a crucial role in achieving field production management and ensuring food security.In this study,RGB images and hyperspectral data were acquired using a UAV with visible and hyperspectral cameras at three fertility stages of wheat,and field measurements were also collected.By extracting feature parameters from the UAV image data and conducting correlation analysis and variable screening,the optimal combination of feature parameters screened at each stage was selected for the estimation of wheat agronomic parameters in this study.In order to comprehensively characterise the growth of wheat,three agronomic parameters,namely above-ground biomass,chlorophyll content and leaf nitrogen content,were combined to construct a comprehensive growth index,and further studies were carried out to monitor the growth of wheat.On this basis,the integrated growth index was introduced for analysis and an optimal model for wheat yield estimation was proposed in order to accurately and efficiently estimate wheat yield.The main conclusions are as follows:(1)This study extracted feature parameters from UAV image data and conducted correlation analysis with various agronomic parameters.Variable selection was performed using the CARS algorithm to construct estimation models for each agronomic parameter.The results showed that the optimal model for wheat above-ground biomass during the jointing stage was the support vector regression(SVR)model,with an R2 of 0.72 and an RMSE of 214.33 kg/hm2.The optimal model during the booting stage was the Gaussian process regression(GPR)model,with an R2 of 0.73 and an RMSE of 256.55 kg/hm2.The optimal model during the flowering stage was also the GPR model,with an R2 of 0.71 and an RMSE of 329.82 kg/hm2.For wheat chlorophyll content,the optimal model during the jointing stage was the SVR model,with an R2 of 0.71 and an RMSE of 0.361.The optimal model during the booting stage was the multiple stepwise regression(MLR)model,with an R2 of 0.67 and an RMSE of 0.272.The optimal model during the flowering stage was the GPR model,with an R2 of 0.74 and an RMSE of 0.311.For wheat leaf nitrogen content,the optimal model during the jointing stage was the GPR model,with an R2 of 0.74 and an RMSE of 0.662g/kg.The optimal model during the booting stage was the MLR model,with an R2 of 0.68 and an RMSE of 0.690g/kg.The optimal model during the flowering stage was the GPR model,with an R2 of 0.69 and an RMSE of 0.625g/kg.(2)This study introduces the equal weighting method and coefficient of variation method to construct a new comprehensive growth index(CGI)to overcome the limitations of a single index in characterizing crop growth.The constructed CGI and feature parameters extracted from unmanned aerial vehicle(UAV)images were subjected to correlation analysis and variable selection.A comprehensive growth index estimation model was constructed using the selected parameter combination.The results showed that for estimating CGIavg,the optimal model for jointing stage was the SVR model with R2 of 0.77 and RMSE of 0.095,the optimal model for booting stage was the GPR model with R2 of 0.71 and RMSE of 0.098,and the optimal model for flowering stage was the SVR model with R2 of 0.78 and RMSE of 0.087.For estimating CGIcv,the optimal model for jointing stage was the MLR model with R2 of 0.73 and RMSE of 0.084,the optimal model for booting stage was the GPR model with R2 of 0.74 and RMSE of 0.092,and the optimal model for flowering stage was the SVR model with R2 of 0.78 and RMSE of 0.085.(3)This study explored the correlation between wheat yield and the feature parameters extracted from unmanned aerial vehicle(UAV)data,and compared the effects of using a comprehensive growth index(CGI)to estimate yield.The accuracy of direct yield estimation using feature parameters was found to be lower than that of yield estimation using the optimal parameter combination based on the estimated CGI.By comparing the yield estimation methods,it was found that the yield estimation model based on the optimal parameter combination of CGIcv had the highest accuracy,with the GPR model being optimal during the flowering stage,with an R2 of 0.68 and an RMSE of 258.35 kg/hm~2.