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

Quinoa, recognized as a nutritionally dense and functionally versatile food source, necessitates non-destructive approaches for quality attribute assessment. This investigation demonstrates enhanced predictive accuracy for three critical quality parameters-protein, starch, and total phenolic content-through spectral information fusion (VIS-NIR and NIR) and spectral feature augmentation, thereby validating the feasibility of employing spectral fusion and enhancement technologies for quinoa quality evaluation. By intensifying characteristic bands closely associated with quinoa quality attributes (identified through Random Frog, Successive Projections Algorithm, and Competitive Adaptive Reweighted Sampling), the spectral model performance was significantly improved(Protein: R_P²=0.9561, RMSEP = 0.1924;Starch:R_P²=0.9692，RMSEP = 0.2175;Total phenols:R_P²=0.9420, RMSEP = 0.0325). Furthermore, an innovative framework integrating explainable artificial intelligence with optimal predictive modeling was developed to enhance quinoa quality assessment. The model's validity and interpretability were substantiated through Shapley Additive exPlanations (SHAP) values. Collectively, the proposed spectral information supplementation methodology and novel model evaluation strategy provide significant insights for quinoa quality attribute detection.