Hyperspectral data-driven corn nitrogen monitoring: application and interpretability analysis of multi-source feature...

Area of Science:

• Agricultural Science
• Remote Sensing
• Data Science

Background:

• Accurate monitoring of canopy nitrogen content is vital for optimizing maize production and minimizing environmental impact.
• Existing hyperspectral data models face challenges in accuracy and interpretability, limiting practical application in industrial maize farming.

Purpose of the Study:

• To develop an advanced framework for accurate and interpretable monitoring of canopy nitrogen content in maize using hyperspectral data.
• To optimize spectral band selection and construct novel spectral indices for enhanced nitrogen estimation.

Main Methods:

• Compared Genetic Algorithm (GA) and Successive Projections Algorithm (SPA) for spectral band optimization.
• Developed 0-2 order fractional-order derivative (FOD) based 2D and 3D spectral indices.
• Constructed a stacked ensemble learning model using XGBoost, GBDT, and Bayesian Ridge, incorporating interpretability techniques.

Main Results:

• The GA-SPA hybrid strategy enhanced spectral band selection.
• 3D spectral indices derived from FOD outperformed traditional vegetation and 2D indices (R²p = 0.801).
• The stacked ensemble model with optimized features achieved the highest accuracy (R²p = 0.826), with red-edge and near-infrared regions being key contributors.

Conclusions:

• The integrated framework offers a robust and interpretable solution for precise maize nitrogen management.
• This approach enhances crop production efficiency while mitigating environmental consequences associated with nitrogen use.