Application of hyperspectral imaging for detecting and visualizing leaf lard adulteration in minced pork

Abstract

The present study describes a methodology using hyperspectral imaging (HSI) for detecting and visualizing the illegally substituted or mislabeled homologous fats of leaf lard adulteration in minced pork non-destructively and rapidly. Hyperspectral images (400–1000 nm) of 176 adulterated samples with adulteration percentages of 0–100% (w/w) at 10% increments were scanned. Different mathematical preprocessings were individually applied to the average spectra extracted from regions of interest (ROIs), and quantitative calibration models were constructed using principal component analysis (PCR) and partial least squares regression (PLSR), respectively. Subsequently, PLSR model developed by second-order derivative preprocessed full spectra performed best and was finally chosen. After that, four groups of wavelengths were selected using principal component (PC) loadings, two-dimensional correlation spectroscopy (2D-COS), competitive adaptive reweighted sampling (CARS) and regression coefficients (RC), respectively. The preferred RC-PLSR model was emerged with the Rp2 of 0.98, root mean square error (RMSEP) of 4.87%, residual predictive deviation (RPD) of 6.57, and the limit of detection (LOD) of 6.08% in prediction set. Adulteration maps were successfully generated by calculating the predicted adulteration response at each pixel, which could not be accomplished by the naked eye. Control samples with known distributed patterns were also visualized, and the results were consistent with actual situation to make the validity proven. In the light of the findings, overall results denote that HSI combined with multivariate analysis possesses the ability to detect and visualize pork fats adulterated in minced pork.