Estimating Maize Leaf Area Index Using Multi-Source Features Derived from UAV Multispectral Imagery and Machine...

Area of Science:

• Agricultural Remote Sensing
• Plant Physiology
• Machine Learning in Agriculture

Background:

• Leaf Area Index (LAI) is crucial for assessing crop health and yield.
• UAV multispectral imagery offers rich data but has limitations in LAI estimation using single features.
• Accurate LAI estimation is vital for precision agriculture and effective crop management.

Purpose of the Study:

• To develop and evaluate a multi-source feature fusion framework for estimating maize LAI using UAV multispectral imagery.
• To integrate vegetation indices (VIs), texture features (TFs), and texture indices (TIs) for enhanced LAI estimation.
• To assess the performance of stacked ensemble machine learning models (PLSR, SVM, RF, GBDT) for maize LAI prediction.

Main Methods:

• Conducted field experiments with varying maize planting densities and nitrogen rates.
• Acquired UAV multispectral imagery to extract VIs, TFs, and TIs.
• Employed a stacked ensemble approach combining PLSR with SVM, RF, and GBDT algorithms for feature fusion and LAI estimation.

Main Results:

• The integrated framework significantly improved LAI estimation accuracy compared to using VIs alone.
• Fusion of VIs, TFs, and TIs with PLSR+GBDT achieved the highest R² (0.844) and lowest RMSE (0.436).
• Independent validation confirmed the robustness of the multi-model fusion framework (PLSR+GBDT) with R² values of 0.859 and 0.794.

Conclusions:

• Multi-source feature integration using machine learning enhances the accuracy and robustness of maize LAI estimation.
• The developed framework provides a valuable tool for precision agriculture and real-time crop growth monitoring.
• This approach overcomes limitations of single-feature analysis in remote sensing-based crop assessment.