Advancing quinoa(Chenopodium quinoa Willd.) quality assessment using hyperspectral imaging

Area of Science:

• Agricultural Science
• Food Science
• Analytical Chemistry

Background:

• Quinoa is a nutrient-dense food requiring non-destructive quality assessment methods.
• Accurate measurement of protein, starch, and phenolic content is crucial for quinoa quality.

Purpose of the Study:

• To enhance predictive accuracy for key quinoa quality attributes using spectral techniques.
• To validate spectral fusion and feature augmentation for improved quinoa quality evaluation.
• To develop an interpretable AI model for quinoa quality assessment.

Main Methods:

• Combined Visible Near-Infrared (VIS-NIR) and Near-Infrared (NIR) spectral data.
• Employed spectral feature selection algorithms (Random Frog, SPA, CARS) for band selection.
• Integrated explainable artificial intelligence (Shapley Additive exPlanations - SHAP) with predictive models.

Main Results:

• Significantly improved prediction models for protein (R_P²=0.9561), starch (R_P²=0.9692), and total phenols (R_P²=0.9420).
• Spectral feature augmentation intensified characteristic bands, boosting model performance.
• SHAP values confirmed the validity and interpretability of the developed AI model.

Conclusions:

• Spectral information fusion and feature enhancement offer a robust approach for non-destructive quinoa quality assessment.
• The integrated AI framework provides reliable and interpretable insights into quinoa quality attributes.
• This methodology advances the detection of critical quality parameters in quinoa.