A novel NIR spectral calibration method: Sparse coefficients wavelength selection and regression (SCWR).

Area of Science:

• Chemometrics
• Spectroscopy
• Data Analysis

Background:

• Near-infrared (NIR) spectroscopy is widely used for chemical analysis.
• Simultaneous regression and wavelength selection are crucial for efficient NIR data analysis.
• Existing methods often involve random procedures or cross-validation, increasing complexity and time.

Purpose of the Study:

• To propose a novel sparse coefficients wavelength selection and regression (SCWR) method.
• To enable rapid and simultaneous regression and wavelength selection on NIR datasets.
• To develop variants for selecting a specified number of wavelengths.

Main Methods:

• Formulating spectral calibration as a least absolute shrinkage and selection operator (LASSO) problem.
• Reformulating the problem into an alternative direction multiplier method (ADMM) framework.
• Developing sparse coefficients wavelength selection (SCWS) and specified number SCWR (NSCWR) methods.

Main Results:

• SCWR methods demonstrated superior performance on potato, corn, and soil NIR datasets.
• Fewer feature wavelengths were selected compared to existing simultaneous methods.
• SCWR-based methods identified wavelengths with higher information content for prediction.
• Regression performance showed robustness to hyperparameter variations within proper ranges.

Conclusions:

• SCWR offers an efficient and effective approach for simultaneous regression and wavelength selection in NIR spectroscopy.
• The method provides improved predictive accuracy with parsimonious models.
• SCWR is a valuable tool for analyzing complex NIR datasets across various applications.