Application And Assessment Of UAV On-board Multispectral Sensor For Non-destructive Site-specific Rapeseed Crop(brassica Napus L.) Phenotype Variable And Weed Discrimination At Early Phenological Stage

The ever-increasing population and urbanization lead to the decreased availability of the total cultivated land,which in addition to the climate changes,is putting greater pressure on the agriculture sector.Conventional agriculture practices and lack of information about crop variations lead to low-quality products and environmentally harmful substances.Therefore,precision agriculture(PA)is an environment-friendly technique used to determine,analyze and manage the factors to maximize production,profit and mediate the adverse effects on the environment.Among the many approaches applied in PA,sensors on-board unmanned aerial vehicle(UAV)combined with geospatial instruments detect crop variability and offer a precise and reliable solution in real-time for crop monitoring and management.This dissertation focuses on unmanned aircraft vehicle(UAV)based remote sensing for the non-destructive estimation of functional traits such as leaf area index(LAI),leaf mass per area(LMA),and specific leaf area(SLA)of rapeseed crop at different physiological stages as well as for early season weed discrimination and mapping.The present study mainly aims to(i)evaluate the potential of UAV on-board multispectral sensor in relation to crop biophysical parameters,such as LAI and DW,during the growing season of rapeseed crops at different phenological stages,(ii)assess the spectrally predicated leaf DW and LAI for the calculation of LMA and SLA at different phenological stages and resolutions(iii)evaluate and determine the optimum vegetation indices(VIs)for early season weed detection and discrimination over oil rapeseed crop(iv)to compare the pixel-based and object-based image analysis methods for weed discrimination and site-specific weed infestation mapping.The oil rapeseed crop at the experimental stations of Zishi,Jingzhou,and Huazhong Agricultural University,Wuhan,China,was selected as the trail fields.The main results of this study are as follow:(1)A wide range of variations in LAI,DW,LMA,and SLA due to differences in canopy and leaf sizes/shapes at different phenological stages were observed.The VIs investigated in this study include RVI,NDVI,GNDVI,BNDVI,SAVI,OSAVI,MSAVI,MSAVI2,and MTVI2 at different growth stages revealed close associations with the crop reflectance characteristics.The coefficient of determination(R~2)and root mean square error results indicated the significant accuracy of LAI and DW estimation.RVI and MTVI2confirmed optimum VIs for LAI estimation at all phenological stages(seedling,elongation,and flowering).They attained the highest R~2 and lowest RMSE values by RVI(R~2=0.93,RMSE=0.30)and MTVI2(R~2=0.89,RMSE=0.38)at the elongation stage wherein RVI,NDVI and GNDVI demonstrated optimum VIs for DW at all phenological stages.The highest R~2 and lowest RMSE values were attained by RVI(R~2=0.75,RMSE=0.27)and GNDVI(R~2=0.75,RMSE=0.26),followed by NDVI(R~2=0.70,RMSE=0.29)at the elongation stage.The noise equivalent of sensitivity and uncertainty analysis accounted for the sensitivity of VIs,which revealed the optimal VIs of RVI and MTVI2 in the LAI estimation.(2)The potential of multispectral VIs for oil rapeseed crop LAI and DW estimation was confirmed at all growth stages,except the maturity stage,and the empirically predicted LAI and DW was found critical and accurate for the estimation of LMA and SLA,which showed significant accuracies at(R~2=0.85,RMSE=26.3)and(R~2=0.81,RMSE=0.001)at the elongation stage followed by(R~2=0.85,RMSE=23.4)and(R~2=0.71,RMSE=0.001)at flowering stage,respectively.Therefore,we conclude that VIs based developed models can reliably estimate phenotype variables at phenological stages of oil rapeseed crop.(3)In terms of image resolutions,the raw image data with ground resolution(20 cm)were resampled to 30,50,and 100 cm by using the pixel aggregate interpolation method to assess the influence of resolution on the estimation of phenotype variables by using VIs.However,robust results can still be obtained when the imagery of ground resolution of100cm was used for oil rapeseed crop characterisation.Therefore,high-altitude images will be preferred in obtaining a decreased number of images,which will significantly influence image acquisition and processing time.(4)The UAV on-board multispectral camera(Mica Sense Red Edge)was helpful in demarcating the early-season weed detection,and discrimination and OBIA methods showed superior performance over the pixel-based method for site-specific weed discrimination and mapping.Spectral analysis using vegetation indices,Kmeans clustering algorithm,Fusion of VI and Kmeans,support vector machine(SVM),and artificial neural network(ANN)modelling methods efficiently determined weed infestation and discrimination,where ANN and SVM obtained an overall accuracy(OA=0.98%),and Kappa coefficient(KC=0.96),followed by Fusion of VI and Kmeans with OA=0.93%and KC=0.88.The study formulates a precise method for identifying and estimating weed management zones,which is crucial for site-specific weed management.This study first reported the potential of empirically predicted LAI and DW to estimate LMA and SLA at different growth stages of rapeseed crops.Multispectral VIs based newly developed dynamic models can reliably estimate LAI,LMA,and SLA at the three phenological stages.The multispectral VIs were potentially crucial in identifying the sensitive spectral region for weed detection and discrimination at the early phenological stage.In brief,this study successfully developed a deeper understanding for the estimation of biophysical characteristics of rapeseed crops and can be well extended to other crops of higher importance and sensitivity to weeds and climatic variations using PA practices.