Construction Of A Model For Monitoring The Growth Of Maize During Critical Fertility Periods Based On UAV Images

Crop growth monitoring is very important for guiding field management and crop yield estimation.In this study,based on the different sensitivity of canopy spectral reflectance to crop agronomic parameters during different growth periods of maize,a field experiment with different levels of nitrogen application was carried out.The visible light camera on the DJI UAV platform was used to acquire canopy remote sensing images of maize at the seedling stage,nodulation stage,trumpeting stage,tasseling stage,filling stage and finishing stage,and to collect and analyze field data of maize leaf area index,leaf nitrogen concentration and above-ground biomass simultaneously.Pre-processing of UAV visible light by image stitching and image masking,constructing image feature parameters and maize growth parameters models,and screening optimal image feature parameters and optimal fertility period.Univariate regression models,partial least squares regression models and BP neural network models based on image feature parameters were developed and the results of the study are as follows:1.The leaf N concentration showed a decreasing trend in the whole fertility period of maize under different N application levels,and the leaf N concentration increased with the increase of N application in all periods;the image characteristic parameters had the highest correlation with leaf N concentration at the nodulation stage of maize,and the parameters NGI and NGBDI had higher correlation with leaf N concentration,with correlation coefficients of 0.89 and 0.77,respectively;among the four UR,exponential,linear and power function models,the power function model with the image feature parameter NGI as the independent variable had the best inversion,with a coefficient of determination of 0.725.2.The above-ground biomass of maize showed an increasing trend during the whole reproductive period,with N5>N4>N3>N2>N1>N0 at each reproductive period from nodulation to filling,and N4>N5>N3>N2>N1>N0 at maturity;the image parameters were the most correlated with above-ground biomass at tassel stage,among which the characteristic parameters B/G,NGBDI,NGI and G/R were significantly correlated with above-ground biomass.The power function regression model of image parameters G/R had the best effect with a coefficient of determination of 0.639;the PLSR model had the best inverse effect with the combination of NGRDI,R/G,G-B and NRI with a coefficient of determination of 0.681;the BPNN had the best inverse effect with the combination of NGRDI,R/G,G-B and NRI with a coefficient of determination of 0.744.is 0.744.3.The LAI tended to increase and then decrease throughout the reproductive period,and the LAI increased with the increase of nitrogen application at each reproductive period;the image feature parameters at the male tapping stage were best correlated with LAI,and the image feature parameters NGI,EXG,NGBDI and NPCI were significantly or highly significantly correlated with LAI at each reproductive period,and the coefficient of variation was small;the power function model had the best inversion effect with EXG,and the coefficient of determination was 0.642.The inversion of the PLSR model with the combination of the feature parameters EXG,RGBVI,MGRVI and GRVI was the best,with an R2 of 0.683;the inversion of the BPNN model with the combination of the feature parameters EXG,RGBVI,MGRVI and GRVI was the best,with a coefficient of determination of 0.794.The inversion of the BPNN model was the best among the three different modeling methods.