A three-stage wavelength selection algorithm for near-infrared spectroscopy calibration.

Area of Science:

• Analytical Chemistry
• Chemometrics
• Spectroscopy

Background:

• Near-infrared (NIR) spectral data is high-dimensional and often contains redundant information.
• Existing wavelength selection methods struggle with collinearity, impacting model accuracy and reliability.
• Effective wavelength selection is crucial for developing robust predictive models in spectroscopy.

Purpose of the Study:

• To propose and evaluate a novel three-stage wavelength selection algorithm (Stage III) for NIR spectral data.
• To reduce data redundancy and inter-wavelength collinearity in NIR spectroscopy.
• To develop a simpler and more accurate predictive model for corn sample analysis.

Main Methods:

• A three-stage algorithm involving correlation analysis and stepwise regression was developed.
• Stage I: Selects wavelengths with high correlation to the concentration vector.
• Stage II: Selects wavelengths with low inter-correlation.
• Stage III: Utilizes stepwise regression to identify significant wavelengths for modeling.
• A multiple linear regression (MLR) model was developed using the selected wavelengths.

Main Results:

• The Stage III-MLR model demonstrated superior performance compared to full spectrum, Stage I, and Stage II models.
• The Stage III-MLR model achieved a high coefficient of determination (0.9360) on the test set.
• Stage III outperformed established methods like SPA, UVE, and CARS in prediction accuracy.

Conclusions:

• The proposed three-stage wavelength selection algorithm is effective for NIR spectral data analysis.
• Stage III successfully reduces collinearity and simplifies model complexity, enhancing prediction precision.
• This algorithm offers an effective approach for modeling NIR spectroscopy with improved accuracy.