Growth Estimation And Nitrogen Nutrition Diagnosis Of Winter Wheat Based On UAV Multi-spectral Remote Sensing

Wheat is an important food crop in China,and providing an important material basis to build a moderately prosperous society in all aspects for China.However,in the process of wheat cultivation,there is also a widespread over-application of nitrogen,resulting in waste of resources and agricultural surface pollution.It is important to conduct research on monitoring crop growth and nitrogen nutrient status to achieve precision management of farmland with the goal of high yield and efficiency of wheat in large fields.In this study,a nitrogen application field experiment of winter wheat was carried out from 2020 to 2022.Wheat variety Xiaoyan 22 was used as the test material,and four nitrogen application levels were set up:120,150,180 and 210kg N/hm²,each treatment was repeat eight times.The spectral feature information of the acquired UAV multispectral images was combined with the growth parameters such as leaf area index,above-ground biomass,leaf nitrogen content,and spike nitrogen content measured on the ground to establish a model of wheat growth parameters.Based on this,the paper further investigated the estimation of yield and N-nutrient diagnosis methods based on the growth parameters modeled by the model simulation,aiming to provide theoretical basis and technical support for accurate monitoring and management of wheat in farmland.The main research work and results of the paper are as follows:（1）The selection of vegetation indices acquired by UAV multispectral imagery is difficult due to the many combinations of vegetation indices.This paper proposes an improved hybrid coding method that enables simultaneous model parameter optimization and feature screening.These hybrid coding approaches build a coupled machine learning model by improving the particle swarm algorithm and the gray wolf particle swarm algorithm.The results showed that the optimal combination of spectral features to simulate the LAI changes during the whole reproductive period was G,DVI,MSR and RVI.Its driven the CGWOPSO-SVM model is the most suitable model in this study with validation period of RMSE=0.219,R²=0.917,MAE=0.161,and NRMSE=0.053.The optimal combination of spectral features to simulate changes in above-ground biomass during the full reproductive period was MREDVI,NDVI＿RVI and NGI,which drove the CGWOPSO-SVM model,in this study,is the most suitable model,which validated period with RMSE=0.251 kg/m²,R²=0.839,MAE=0.199and NRMSE=0.211 kg/m².（2）The leaf N content is much higher than the spike N content during the reproductive growth period of wheat,but it is difficult to distinguish each other when the wheat canopy is observed by UAV remote sensing.To address this problem,this paper introduces the leaf-spike projection area ratio parameter as a parameter of the coupled machine learning model to improve the simulation accuracy of the canopy N content.The results showed that the leaf-spike projection area ratio was imported as a parameter into the CGWOPSO-XGB model,and it was found that the accuracy of the canopy nitrogen content model was significantly improved for a leaf-spike area ratio of about 7:3 with the input feature combinations RE,MEVI,REDVI and REWDRVI,and the validation accuracy RMSE=0.258%,R²=0.909,MAE=0.217%and NRMSE=0.086.Compared with the CGWOPSO-XGB model without considering the influence of ear,the error of all growth period decreased by about 15%,and period the error decreased by about 35%during the reproductive growth.（3）To establish a yield estimation model based on the difference of leaf area index and above-ground biomass growth in different periods of UAV inversion through scientific assumptions and theoretical derivation for the current situation that UAV remote sensing inversion of wheat yield mainly relies on empirical methods.The results showed that the model with the highest accuracy of yield estimation at the flowering and maturity stages had a validation period of RMSE=398kg/hm²,R²=0.762,MAE=332kg/hm²,and NRMSE=0.05,the yield model of growth difference between maturity stage and flowering stage model is written as:Y=a(AGB_Maturity-AGB_Flowering)+b(LAI_Maturity-LAI_Flowering)+?.（4）Due to the different errors in the UAV inversion of different growth information of wheat,this information may cause error amplification problems when brought into the critical N concentration dilution model.In this paper,by comparing the critical nitrogen concentration dilution curve model based on wheat leaf area index and the critical nitrogen concentration dilution curve model based on above-ground biomass,a suitable model for nitrogen nutrient diagnosis of unmanned wheat was selected.The results showed that the overall critical N concentration dilution curve model based on the leaf area index was somewhat misclassified and basically diagnosed as excess N nutrition.The critical N concentration dilution curve model based on above-ground biomass for N nutrition diagnosis was better,and its conformity with the law of fertilization and follow-up was the recommended method in this study.In summary,the coupled machine learning model established in this paper can determine the key vegetation characteristics of simulated growth and reveal the relationship between vegetation characteristics and growth information.The combination of the parameter leaf-spike projection area ratio and coupled machine learning model can improve the understanding of spatial distribution of canopy nitrogen concentration.The yield model can quantitatively describe the relationship between aboveground biomass and leaf area index and yield.The above research confirms the high accuracy and practicability of the established model.