基于不同飞行高度无人机多光谱及集成学习的冬小麦叶面积指数反演研究OA

The leaf area index(LAI)is a key biophysical parameter characterizing the canopy structure and growth status of winter wheat,playing a crucial role in precision agriculture monitoring and crop condition assessment.However,existing inversion methods suffer from unstable accuracy and insufficient model generalization capabilities across varying flight altitudes.To adress these limitaions,this study develops an LAI estimation framework based on multi-altitude UAV multispectral data by integrating ensemble learning architectures with multiple machine learning algorithms.Analysis of variance(ANOVA)was applied to analyze the mechanisms through which flight altitude influences spectral reflectance and textural features.Following the extraction of vegetation and textural indices from the multispectral data,Pearson and Spearman correlation analyses were conducted to screen LAI-sensitive features for predictive modeling.Four distinct models—XGBoost,RF,XGBoost-EN,and RF-Ridge—were subsequently established to perform multi-altitude and multi-feature LAI inversions and comparative accuracy evaluations.The results indicate that UAV flight altitude exerts a significant scale effect on both reflectance and textural characteristics,with 80 m indentified as the critical threshold for reflectance variations.Regardless of whether single-feature or mixed-feature inputs were utilized,20 m flight altitude consistently yielded the highest estimation accuracy across all tested altitudes,R2 achieving a maximum of 0.745,RMSE of 0.699,and MAE of 0.507.Furthermore,The XGB-EN and RF-Ridge ensemble models outperformed the standalone machine learning models,exhibiting superior inversion performance with an R2 of up to 0.802,an RMSE down to 0.592,and an MAE down to 0.421,indicating excellent fitting accuracy and stability.Overall,XGBoost-EN is the optimal model for the LAI inversion task.This study validated the feasibility of ensemble learning models and machine learning methods for high-precision LAI inversion across multiple altitudes,providing a theoretical foundation and technical pathway for remote sensing-driven crop growth monitoring.