The Effect Of Tassels On LAI And FAPAR Estimation In Maize

Maize,as one of the three major food crops,has a critical impact on world food security in terms of its yield.Leaf area index(LAI)and fraction of asorbed potosynthetically ative rdiation(FAPAR)are important biophysical parameters for measuring crop growth and are highly correlated with crop yield,quality and other traits.The study of tassels spectra at the canopy level will help to improve the accuracy of estimates of maize LAI and FAPAR,and provide guidance to regional managers in developing more rational plans.Summer maize is the object of this study.This study analyses the changes in canopy vegetation indices based on RGB and multispectral data extracted before and after tassel removal using hand removal and image processing methods to remove tassels,and analyses its effect on the estimated LAI and FAPAR.The main findings of this thesis are as follows.(1)The Unet model has the highest accuracy(PA = 82.14%,Mean IOU = 84.43%)among segmentation methods such as the maximum between-class variance method(OTSU),machine learning methods(RFR)and deep learning models(Deep Lab V3+,Pspnet and Unet)based on RGB data.(2)The effect of the tassels on the RGB vegetation index of the canopy differs between the varieties,with the RGB vegetation index(r,EXR,rg)decreasing and the RGB vegetation index(g,EXG,EXGR,grb,MGRVI,RGBVI,GRVI,GLA,VARI)increasing after the removal of the tassels.(3)Among the different coding structures(VGG,Resnet and Mobilenet)and prediction methods(bilinear interpolation and sliding window prediction)based on the multispectral data Unet,the VGG coding structure combined with the sliding window prediction method had the highest accuracy(Mean IOU=79.68%,PA=82.45%).In the 11 multispectral datasets constructed in this study,the 5Band dataset had the highest accuracy(Mean IOU=89.53%,PA=85.97%).(4)Canopy spectra were decreased by removing the tassels in the multispectral data.The contribution of the tassels to the vegetation index varied between varieties,with the contribution varying with the growth period.Vegetation indices(NDVI,RVI1,RVI2,MSR,GNDVI and NDVI＿R)increased with tassels removed and vegetation indices(DVI,EVI,SAVI,MSAVI2,TVI,MTVI2 and TCARI)decreased with tassels removed.(5)The best prediction accuracy for LAI and FAPAR was obtained using the GBDT method based on vegetation indices extracted after image segmentation to remove tassels(LAI: R2=0.816,RMSE=0.399,rRMSE=0.074;FAPAR: R2=0.802,RMSE=0.025,rRMSE=0.016).The effect of the numbers of vegetation indices on model accuracy was tested based on the contribution of tassels to vegetation indices,and the results showed that 9 VIs predicted LAI with the best accuracy(R2=0.8238,RMSE=0.401,rRMSE=0.109);11 VIs predicted FAPAR with the best accuracy(R2=0.802,RMSE=0.025,rRMSE= 0.027).