Abstract
Leaf area index (LAI) is a critical indicator of canopy architecture and physiological performance, serving as a key parameter for crop growth monitoring and management. Although UAV multispectral imagery provides rich spectral and spatial information, the limitations of single texture features for LAI estimation still require further exploration. To address this issue, this study developed a multi-source feature fusion framework that integrates vegetation indices (VIs), texture features (TFs), and texture indices (TIs) within a stacked ensemble approach combining Partial Least Squares Regression (PLSR) with Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting Decision Tree (GBDT) algorithms to estimate maize LAI.A field experiment was conducted under three planting densities (42,000, 63,000, and 84,000 plants ha−1) and four nitrogen rates (0, 80, 160, 240 kg N ha−1) to assess the potential of UAV-based multispectral imagery for maize LAI estimation. The results show that when using partial least squares regression (PLSR) combined with RF, SVM and GBDT to estimate maize LAI, the R2 values are 0.653, 0.697 and 0.634, and the RMSE is 0.650, 0.608 and 0.668, respectively, when only vegetation indices (VIs) is used as input. After texture features (TFs) incorporation, the R2 increases to 0.717, 0.794, and 0.801, and the RMSE decreases to 0.587, 0.500, and 0.492. Further inclusion of the texture indices (TIs) raises the R2 to 0.789, 0.804, and 0.844, with RMSE of 0.506, 0.489, and 0.436, respectively. Independent test set validation under contrasting conditions confirmed that our multi-model fusion framework (PLSR+GBDT) with multi-source feature fusion (VIs+TFs+TIs) effectively estimated LAI, achieving an R2 of 0.859 and 0.794. These results demonstrate that multi-source feature integration via machine learning enables robust and accurate estimation of maize LAI, providing a valuable tool for precision agriculture and crop growth monitoring.