Estimation of Nitrogen Content in Alfalfa Plants Based on Multi-Source Feature Fusion

Area of Science:

• Agricultural Science
• Remote Sensing
• Plant Physiology

Background:

• Plant nitrogen content (PNC) is vital for assessing crop nutrition and optimizing fertilizer use.
• Accurate PNC monitoring aids in diagnosing crop health, improving fertilizer efficiency, and minimizing agricultural pollution.

Purpose of the Study:

• To evaluate the effectiveness of multispectral vegetation indices (VIs) and texture indices (TIs) derived from UAV imagery for estimating alfalfa PNC.
• To compare the performance of four machine learning models (RFR, SVR, BPNN, XG-Boost) in predicting alfalfa PNC using VIs and TIs.

Main Methods:

• Extracted VIs and texture feature values (TFVs) from UAV multispectral imagery during critical alfalfa growth stages.
• Constructed TIs by combining TFVs and identified highly correlated variables with PNC.
• Trained and validated four machine learning models using VIs, TIs, and their combined features for PNC estimation.

Main Results:

• TIs showed stronger correlations with alfalfa PNC (|r| > 0.6) than raw texture values.
• Integrating VIs and TIs significantly improved PNC estimation accuracy across growth stages (R² increased by 5.4-19.7%).
• The XG-Boost model with combined VIs and TIs achieved the highest accuracy, with validation R² of 0.80 during the budding stage.

Conclusions:

• Integrating multispectral VIs and TIs derived from UAV imagery is effective for accurate alfalfa PNC estimation.
• This approach provides valuable scientific support for precision field management and fertilization strategies in alfalfa cultivation.
• Enhanced PNC monitoring contributes to efficient nitrogen management and reduced environmental impact.