Estimating the rice nitrogen nutrition index based on hyperspectral transform technology

Area of Science:

• Agricultural Science
• Remote Sensing
• Plant Nutrition

Background:

• Rapid diagnosis of rice nitrogen nutrition is crucial for effective field management and precision fertilization.
• The nitrogen nutrition index (NNI) is a key parameter for quantitative diagnosis, but current hyperspectral methods often overlook the relationship between critical nitrogen concentration and spectral reflectance.

Purpose of the Study:

• To develop an improved method for estimating the nitrogen nutrition index (NNI) in rice using hyperspectral remote sensing.
• To investigate the impact of spectral data transformations on NNI estimation accuracy.

Main Methods:

• Canopy spectral data acquired via UAV hyperspectral remote sensing.
• Determination of rice critical nitrogen concentration curve and NNI.
• Application of logarithmic difference transformation and Autoencoder for feature extraction.
• Construction of NNI inversion models using Extreme Learning Machine (ELM) and Bald Eagle Search-Extreme Learning Machine (BES-ELM).

Main Results:

• Logarithmic difference transformation significantly improved NNI estimation compared to simple transformations.
• The BES-ELM model using logarithmic difference spectral features achieved the highest accuracy (R² = 0.839 for training, 0.837 for verification).
• Samples were successfully classified into nitrogen-rich (NNI ≥ 1) and nitrogen-deficient (NNI < 1) groups.

Conclusions:

• Logarithmic difference transformation is an effective approach to enhance NNI estimation accuracy in hyperspectral analysis.
• This study provides a novel method for improving NNI estimation, benefiting precision agriculture and rice management.