Non-destructive authentication of herring powder using near infrared spectroscopy

The increasing demand for herring (Clupea harengus) has heightened the risk of adulteration with non-muscle components such as head and tail parts, necessitating rapid and reliable authentication methods. This study evaluated the potential of near-infrared spectroscopy coupled with chemometric technique for the non-destructive classification and prediction of adulteration in herring powder. Laboratory-prepared samples (0-50% w/w adulteration; head, tail, and combined) and 60 commercial samples were analyzed, yielding 123 samples and 369 spectra. Linear Discriminant Analysis (LDA) achieved high classification performance, with recognition accuracies exceeding 88% and strong discrimination between pure, adulterated, and commercial samples. Classification improved with increasing adulteration levels, reaching near-perfect performance, while lower levels (5-10%) showed reduced sensitivity due to spectral overlap. External validation confirmed good model generalization, with prediction accuracies ranging from 72.05% to 77.52%. Partial Least Squares Regression (PLSR) showed strong calibration performance for adulterant concentration (R² = 0.8913; RMSEC = 5.60% w/w), but moderate predictive ability (R²CV = 0.5739; RMSECV = 11.08% w/w; RPD = 1.54), indicating suitability for screening applications. Among color parameters, b* showed the best predictive performance (R²CV = 0.6984; RMSECV = 1.10; RPD = 1.82), followed by ΔE* (RPD = 1.62), a* (RPD = 1.52), and L* (RPD = 1.50). The findings highlights the applicability of NIR for routine screening while emphasizing the need for enhanced sensitivity for precise quantification at low adulteration levels.