Mining hyperspectral data for non-destructive and rapid prediction of nitrite content in ham sausages

Abstract

Accurate and rapid determination of nitrite contents is an important step for guaranteeing sausage quality. This study attempted to mine hyperspectral data in the range of 900-1700 nm for non-destructive and rapid prediction of nitrite contents in sausages. The average spectra of 156 samples were collected to relate to the measured nitrite values by partial least squares (PLS) regression. Optimal wavelengths were respectively selected by successive projections algorithm (SPA) and regression coefficients (RC) to simplify the PLS model. The results indicated that PLS model established with 15 optimal wavelengths (900.5 nm, 907.1 nm, 908.8 nm, 912.1 nm, 915.4 nm, 920.3 nm, 922.0 nm, 941.7 nm, 979.6 nm, 1083.2 nm, 1213.2 nm, 1353.0 nm, 1460.2 nm, 1595.6 nm and 1699.9 nm) selected by SPA had better performance with rC, rCV, rP of 0.92, 0.89, 0.89 and RMSEC, RMSECV, RMSEP of 0.41 mg/kg, 0.89 mg/kg, 0.49 mg/kg, respectively, for calibration set, cross-validation and prediction set. It was concluded that hyperspectral data could be mined by PLS & SPA for realizing the rapid evaluation of nitrite content in ham sausages.