Ensemble calibration model of near-infrared spectroscopy based on functional data analysis

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Near-infrared spectroscopy (NIRS) is a valuable nondestructive detection technology widely used across diverse fields.
• Advancements in NIRS application necessitate improved data processing techniques for enhanced accuracy and reliability.
• Existing discrete data models often fall short in capturing the continuous nature of spectral data.

Purpose of the Study:

• To propose a novel ensemble calibration model for near-infrared spectroscopy data analysis.
• To introduce a method that addresses the limitations of discrete data models by employing functional data analysis.
• To enhance the robustness and accuracy of NIRS models, particularly for complex sample matrices like soil and agricultural products.

Main Methods:

• The study introduces the functional data analysis-ensemble learning-partial least squares (FDA-EM-PLS) model.
• Near-infrared spectra are divided into intervals, with functional data analysis applied to each segment.
• Sample clustering based on generated functions, Monte Carlo sampling for ensemble learning, and partial least squares regression are utilized.

Main Results:

• The FDA-EM-PLS model effectively reduces noise and provides a basis for variable selection.
• Ensemble learning addresses challenges associated with small sample sizes and improves model robustness.
• Experimental results demonstrate an absolute relative error of less than 10% for both corn and soil data analysis.

Conclusions:

• The FDA-EM-PLS model offers a significant advancement in near-infrared spectroscopy data processing.
• This functional data analysis-based approach enhances model performance and reliability.
• The proposed method shows high potential for accurate quantitative analysis in agricultural and environmental applications.