Nutrition Diagnosis And Yield Prediction Of Winter Wheat Based On Unmanned Aerial Vehicle Multispectral Remote Sensing Under Combined Nitrogen And Phosphorus Application

Winter wheat is one of the main food crops in China,and nitrogen and phosphorus fertilizers have an important impact on the high yield and grain nutrition of winter wheat.LAI,SPAD,and biomass are important indicators to measure the growth status of winter wheat.Therefore,timely and accurate nutrition diagnosis and growth monitoring of winter wheat,as well as reasonable adjustment of fertilizer application to ensure high yield and quality of winter wheat,are of great significance for agricultural production.Remote sensing has been widely used in crop growth monitoring due to its advantages of high efficiency,non-destructive,and large scale.This paper is based on a long-term fixed position nitrogen and phosphorus fertilization experiment of winter wheat summer maize double cropping system in the Guanzhong Plain.During the 2020-2022 winter wheat growth season,winter wheat varieties Pubing 151 and Xinong 979 were used as the research objects.The Dajiang Latitude100 four-rotor Unmanned Aerial Vehicle（UAV）equipped with a Red Edge multispectral camera was used to obtain remote sensing images of winter wheat jointing,booting,flowering,and filling periods,and simultaneously obtain winter wheat nutrients,growth,yield,and other indicators,This paper mainly discusses the calculation and selection of vegetation indices based on multispectral images,which are used to construct winter wheat nutrition diagnosis,growth monitoring,and yield prediction models based on empirical regression and machine learning regression models.Furthermore,the empirical regression model is further applied to the regional scale of Xianyang City,providing theoretical basis and technical support for the realization of large-scale winter wheat nutrition diagnosis,growth monitoring,and yield prediction based on the model established on the unmanned aerial vehicle platform.The main research results obtained are as follows:（1）The nutritional diagnosis results of winter wheat showed that the vegetation index CIgreen and RVI had good correlation with winter wheat plant nitrogen and phosphorus nutritional indicators,respectively.Empirical models constructed using vegetation index and winter wheat nutrition indicators have higher diagnostic accuracy using power functions and quadratic functions,while machine learning regression models using support vector machines and multi-layer perceptron models have better diagnostic effects.Therefore,the diagnostic models for nitrogen content in winter wheat plants constructed based on empirical regression models at jointing,booting,flowering,and filling stages are respectively y=9.7256x^0.557,y=3.10x^0.597,y=4.3687e^0.1133x,y=4.2237x^0.6684,and the diagnostic models for nitrogen accumulation in plants are respectively y=12.27x^0.690,y=3.241x^1.085,y=13.36e^0.195x,y=22.56e^0.292x,and the diagnostic models for phosphorus content in plants are respectively y=0.0006x²+0.003x+2.51,y=11.58x²-7.13x+3.54,y=1.48x²-0.20x+2.26,y=4.2237x^0.6684,and the diagnostic models for plant phosphorus accumulation are y=4.702e^0.010x,y=6.765e^0.014x,y=5.058e^1.976x,and y=5.595x^0.519,respectively.The accuracy of plant nitrogen content in winter wheat nutrition diagnosis is higher,with an accuracy of 75.57%-87.03%.The accuracy of plant phosphorus content in phosphorus nutrition diagnosis is better,with an accuracy of 81.10%-88.80%.（2）The monitoring results of winter wheat growth showed that the correlation between the vegetation index CIgreen and winter wheat LAI,SPAD,and aboveground biomass was good.In empirical regression models,the exponential function and power function constructed by CIgreen were mostly used to monitor winter wheat growth with high accuracy,while in machine learning regression models,the multi-layer perceptron model had a better monitoring effect.The LAI monitoring models for winter wheat based on empirical regression models at jointing,booting,flowering,and filling stages are y=35.80e^0.027x,y=24.96e^0.775x,y=37.96e^0.024x,y=35.36e^0.049x,SPAD monitoring models are y=35.80e^0.027x,y=24.96e^0.775x,y=37.96e^0.024x,y=35.36e^0.049x,and aboveground biomass monitoring models are y=1162e^1.474x,y=1045.7x^0.488,y=3058.4e^0.0817x,respectively,y=3663.4x^0.5473。The monitoring accuracy of aboveground biomass is relatively stable,ranging from 73.55%to 84.91%,and the monitoring accuracy of LAI and SPAD is about 60%or more.（3）The results of winter wheat yield prediction show that RVI,CIgreen,and winter wheat yield have a good correlation.In empirical regression models,power functions constructed by RVI are often used to predict winter wheat yield with high accuracy.In machine learning regression models,random forest models have better monitoring effects,and yield prediction models constructed from the three factors of yield,namely,spike number,grain number per spike,and 1000-grain weight,have higher accuracy.Therefore,the yield prediction model for winter wheat at jointing,booting,flowering,and filling stages constructed based on empirical models is y=1911x^0.3846,y=1156.1x^0.4802,y=2130.3e^0.0911x,y=9060.6x^0.7064,and the prediction accuracy can reach 87.73%-90.89%.（4）The winter wheat growth,nutrition,and yield monitoring models constructed based on vegetation index NDVI were respectively applied to Sentinel-2 satellite imagery to explore the retrieval accuracy and stability of winter wheat growth,nutrition,and yield monitoring models based on the regional scale of Xianyang City.The results show that:Using supervised classification to extract winter wheat planting areas and unmanned aerial vehicle platform to build a model,the monitoring results of winter wheat in Xianyang City are consistent with the actual planting status and growth rules of winter wheat.Therefore,when monitoring regional winter wheat based on satellite images,unmanned aerial vehicle platform can be used to build a model and complete the estimation.At the same time,the process operation of regional scale crop growth monitoring based on unmanned aerial vehicle platform is described,providing theoretical basis and technical support for practical agricultural production applications.