Quantitative Analysis of Soil Total Nitrogen Using Hyperspectral Imaging Technology with Extreme Learning Machine

Area of Science:

• Agricultural Science
• Remote Sensing
• Analytical Chemistry

Background:

• Precise soil nutrient detection is crucial for efficient and sustainable agriculture.
• Total nitrogen (TN) is a key soil nutrient influencing crop yield and environmental health.
• Traditional soil analysis methods can be time-consuming and labor-intensive.

Purpose of the Study:

• To investigate the application of hyperspectral imaging (HSI) for detecting total nitrogen (TN) content in soil.
• To develop and compare chemometric models for accurate TN prediction using spectral data.
• To visualize the spatial distribution of TN in soil samples.

Main Methods:

• Collected 150 soil samples from Lishui City.
• Utilized a hyperspectral imaging (HSI) system in the 900-1700 nm spectral range.
• Extracted characteristic wavelengths using Uninformative Variable Elimination (UVE) and Successive Projections Algorithm (SPA).
• Developed Partial Least Squares (PLS) and Extreme Learning Machine (ELM) calibration models.
• Evaluated model performance using correlation coefficient of prediction (rₚ), RMSEP, and RPD.

Main Results:

• The nonlinear Extreme Learning Machine (ELM) model outperformed the linear Partial Least Squares (PLS) model.
• Models utilizing characteristic wavelengths achieved good prediction results.
• The Uninformative Variable Elimination-Extreme Learning Machine (UVE-ELM) model demonstrated superior performance with rₚ = 0.9408, RMSEP = 0.0075, and RPD = 2.97.
• The UVE-ELM model successfully estimated TN content and generated a distribution map.

Conclusions:

• Hyperspectral imaging (HSI) is a viable technique for the rapid detection and visualization of soil TN content.
• The UVE-ELM model offers a robust and accurate approach for soil TN estimation.
• This research provides a foundation for large-scale soil nutrient monitoring and optimized fertilization practices.