基于无人机多光谱和机器学习的灌溉农田冬小麦产量预测方法研究OA

Accurate prediction of crop yield plays an important role in optimizing planting decisions and improving resource use efficiency.This study focuses on winter wheat in the arid region of northwest Chi-na,and proposes prediction models of vegetation indices for different growth stages such as heading,flow-ering,and grain-filling periods.These models include extreme gradient boosting,support vector regres-sion and random forest regression.The results show that extreme gradient boosting model achieves the highest accuracy in predicting winter wheat yield during the heading and flowering periods,with R2 of 0.92 and 0.90 for the training set,and 0.79 and 0.72 for the test set,respectively.The corresponding combinations of vegetation indices are extra red vegetation index(EXR),modified green-red vegetation index(MGRVI),normalized green-red difference index(NGRDI),red-green ratio index(RGRI),and ratio vegetation index(RVI)for the heading period,and green normalized vegetation index(GND-VI),MGRVI,RVI,EXR,and NGRDI for the flowering period.Random forest regression model achieves the highest accuracy in predicting winter wheat yield during the grain-filling period,with R2 reaching a maximum of 0.81 for the training set,and 0.85 for the test set,indicating that the selected combinations of vegetation indices were all optimal combinations.The experiments show that the prediction accuracy is highest for the grain-filling period,followed by the heading period,and lowest for the flowering period,demonstrating that extreme gradient boosting trees and random forest regression models are applicable to the accurate prediction of winter wheat yield.