Winter Oilseed Rape LAI Inversion via Multi-Source UAV Fusion: A Three-Dimensional Texture and Machine Learning...

Area of Science:

• Agricultural Remote Sensing
• Precision Agriculture
• Crop Monitoring

Background:

• Leaf Area Index (LAI) is crucial for crop growth assessment and management.
• Unmanned aerial vehicle (UAV)-based multispectral data offer potential for LAI estimation.
• Limitations exist in using single texture features for accurate LAI prediction.

Purpose of the Study:

• To explore the efficacy of combining vegetation indices and novel 3D texture features for winter oilseed rape LAI estimation.
• To evaluate the performance of Support Vector Machine (SVM), Backpropagation Neural Network (BPNN), and Extreme Gradient Boosting (XGBoost) models.
• To develop an accurate and reliable method for UAV-based LAI monitoring in precision agriculture.

Main Methods:

• Field experiments collected winter oilseed rape ground truth LAI and UAV multispectral data over two years.
• Vegetation indices and canopy texture features were extracted; novel 3D texture indices were generated using a correlation matrix.
• Significant variables were selected and input into SVM, BPNN, and XGBoost models for LAI estimation.

Main Results:

• Most vegetation indices and texture features showed significant correlations with LAI (p < 0.05).
• The 3D texture index NDTTI demonstrated the highest individual correlation with LAI (R = 0.725).
• The XGBoost model integrating vegetation indices, texture features, and 3D texture indices achieved the highest accuracy (R² = 0.882, RMSE = 0.204).

Conclusions:

• Combining diverse spectral and spatial features significantly improves winter oilseed rape LAI estimation accuracy.
• The developed methodology provides an effective approach for UAV-based multispectral LAI monitoring.
• This research supports scientific and technical advancements in precision agriculture management.