Remote sensing inversion of nitrogen content in silage maize plants based on feature selection

Area of Science:

• Agricultural Science
• Remote Sensing
• Data Science

Background:

• Excessive nitrogen application and low nitrogen use efficiency are significant challenges in China's agriculture.
• Nitrogen is vital for crop growth, yield, and quality, necessitating precise management.

Purpose of the Study:

• To develop accurate canopy nitrogen content inversion models for maize using multispectral remote sensing data.
• To improve nitrogen use efficiency and reduce environmental pollution through precise fertilization.

Main Methods:

• Employed multispectral remote sensing images and field-measured nitrogen content.
• Developed and compared three canopy nitrogen content inversion models: backpropagation neural network (BP), support vector machine (SVM), and partial least squares regression (PLSR).
• Utilized feature selection to eliminate redundant spectral information and identified key spectral indices (GI, NRI) and bands (green, red).

Main Results:

• Feature selection improved modeling efficiency by removing redundant spectral indices.
• Green Index (GI) and Nitrogen Reflectance Index (NRI) showed strong correlations with maize canopy nitrogen content.
• The random forest (RF) algorithm combined with PLSR achieved superior predictive performance, improving accuracy by 3.5%-6.5% over standalone PLSR.

Conclusions:

• Multispectral remote sensing, combined with advanced algorithms like RF-PLSR, enables precise nitrogen diagnosis in maize.
• Optimized nitrogen management strategies can be developed based on accurate canopy nitrogen content estimation.
• This approach provides a scientific basis for sustainable maize cultivation and fertilizer management.