In order to efficiently and accurately predict wheat yield,winter wheat in Zhejiang province was taken as research object.Canopy multi-spectral data at five key growth stages (jointing stage,booting stage,heading stage,grain filling stage,and maturity stage)of winter wheat was obtained by using multispectral sensor onboard the four-rotor Phantom 4 UAV. Five characteristic bands of multispectral camera were selected to calculate 72 vegetation indices at different growth stages. The yield estimation models at different growth stages were constructed by stepwise multiple linear regression (SMLR),partial least squares regression (PLSR),BP neural network (BPNN),support vector machine (SVM) and random forest (RF). Finally,the coefficient of determination (R),root mean square error (RMSE) and relative error (RE) were used to evaluate the estimation model and select the best estimation model. The results showed that the model based on RF has the best estimation effect,and the models based on SMLR,PLSR and SVM have similar estimation effect. The R,RMSE and RE of the models at jointing stage,booting stage,heading stage,grain filling stage and maturity stage were 0.92,0.35,11%; 0.93,0.33,10%; 0.94,0.32,9%; 0.92,0.36,9%; 0.77,0.67,33%,respectively. In the validation of the model,the estimation effect at heading stage was the best (R,RMSE and RE were 0.91,0.35 and 15%,respectively),and the estimation effect at jointing stage,booting stage and grain filling stage was close and had good estimation ability,and the estimation accuracy at maturity stage was the worst (R,RMSE and RE were 0.71,0.47 and 13%,respectively). Therefore,the combination of machine learning algorithm and UAV multispectral extraction of vegetation index can improve the effect of wheat yield estimation. |