UAV hyperspectral imaging enables accurate pre-flowering prediction of faba bean yield and categories. FBR achieved slightly higher estimation accuracy than VIs, further improved by spectral–texture fusion. DNN and XGBoost deliver robust yield regression and yield categories classification across growth stages. Accurate and nondestructive prediction of crop yield and yield categories are crucial for advancing precision. This study presented an integrated framework combining unmanned aerial vehicle (UAV) based hyperspectral imaging and machine learning to predict the early yield and yield categories of faba bean. Field experiments were conducted across three key growth stages—branching, early budding and mid budding—using high-resolution hyperspectral data. Full-band reflectance (FBR) and texture features (TF) were extracted and fused to enhance model sensitivity to canopy spectral-structural variations. The results revealed that FBR achieved higher coefficients of determination ( R 2 >0.60) and lower root mean square errors (RMSE0.86). Spatial analyses confirmed strong consistency between measured and predicted distributions of yield and yield categories, validating the robustness of the proposed approach. In this study, the hyperspectral and machine learning prediction models were developed for early prediction of faba bean yield and yield categories, which provided a scalable data-driven tool for high-throughput phenotypic analysis and sustainable crop management.
Ya et al. (Sun,) studied this question.