Inversion Of Nitrogen Nutrition Parameters Of Spring Maize In Northeast China Based On UAV Hyperspectral Imagery

The unmanned aerial vehicle(UAV)airborne hyperspectral imaging system has important research significance in the field of precision agriculture due to its high spectral resolution,low cost,strong timeliness,and less impact from environmental weather.Precision nitrogen fertilizer management can improve nitrogen use efficiency and reduce environmental pollution while ensuring high yield of spring maize in Northeast China,and has good economic and environmental benefits.Therefore,it is essential to diagnose the nitrogen nutrition status of spring maize effectively in real time.This study is based on the experimental field of Sankeshu Village,Siping City,Jilin Province.The airborne imaging hyperspectral system was used as a means of acquiring hyperspectral data to collect canopy hyperspectral image data of two years of spring maize.The five spectral preprocessing methods were combined with three algorithms(partial least squares regression,BP Neural network and random forest algorithm)inversion of nitrogen nutrition parameters of spring maize,such as aboveground biomass,nitrogen concentration,nitrogen uptake and nitrogen nutrition index.The main work and conclusions are as follows:(1)The best modeling method is the random forest method.The RF random forest algorithm is significantly improved compared with the PLSR algorithm,but the improvement is not obvious compared with the BP neural network.However,the RF algorithm has fast training speed and high efficiency.The RF algorithm have more advantages when diagnose the nitrogen nutrition status of spring corn with multiple samples.(2)Four agronomic indexes(biomass,nitrogen concentration,nitrogen uptake and nitrogen nutrition index)were evaluated by different models to diagnose nitrogen nutrition status of spring maize.After average pretreatment and model accuracy,the predicted R~2 of biomass,nitrogen concentration,nitrogen uptake,and nitrogen nutrition index were 0.56,0.51,0.66,and 0.73,respectively.The accuracy of inversion of nitrogen nutrition index by six processing methods combined with three inversion models is the best,which is conducive to the diagnosis of nitrogen nutrition in the critical growth period of spring maize.(3)Pretreatment of the original hyperspectral can improve the accuracy of the model.The best spectral preprocessing method for this experiment is multi-scatter correction(SMC).The maximum R~2 of the prediction set reached 0.93,0.88,0.89,and0.84,respectively,after averaging the precision of each pretreatment model established by SMC pretreated hyperspectral characteristic parameters and aboveground biomass nitrogen concentration,nitrogen uptake by plants and nitrogen nutrition index.