2023, Vol. 8, Special Issue 5

Yield prediction before harvest is one of the important issues in terms of managing agricultural policies and making the right decisions for the future. The aim of this study is to predict wheat yield using field phenotypic data obtained from unmanned aerial vehicle (UAV) images and advanced machine learning techniques. A UAV platform carrying RGB cameras was employed to collect images of wheat crop. These images (402 images) were combined to form ortho-mosaic image through image processing software Pix4D mapper and were used to extract the vegetation indices (VIs), canopy volume, canopy area by quantum geographic information system (QGIS) open-source software. The yield prediction was done with the help of green leaf area index (GLA), the excess red index (ExR), the excess green index (ExG), the excess green minus excess red index (ExGR), water index (WI), the normalised green-red difference index (NGRDI), the red green blue VI (RGBVI), and the visible atmospherically resistant index (VARI) obtained from UAV RGB images. In addition, some digital variables were also used to reflect the growth trend of wheat, including G/R, G/B, and R/B. The results show that the models can accurately predict yield before the harvest. Support vector machine (SVM) (RMSE=1.025, R2=0.93) and least absolute shrinkage and selection operator (LASSO) regression (RMSE=1.022, R2=0.93) represent the top two best methods for predicting yields among the five typical machine learning models tested in this study. Our findings highlight a potentially powerful tool to predict yield using UAV drone data and advanced machine learning techniques fin other regions and for crops.