Identification of adulterated beef based on near-infrared hyperspectral imaging technique

Abstract. The main objective of this study was to establish a method for improving rapid detection of adulterated beef using hyperspectral imaging technology coupled with chemometrics. Beside 4 minced meat samples of pure beef and 4 of pure duck, 50 Minced beef samples were adulterated with minced duck in the range 2%–50% (w/w) at approximately 2% interval. Hyperspectral imaging was acquired in the reflectance mode. Band threshold was used to extract ROIs and four kinds of partitioning methods (Concentration gradient-CG, Kennard Stone-KS, Sample set Partitioning based on joint x-y distances-SPXY, maximum linear independent method) were used to divide sample into calibration (N=38) and validation (N=12). The spectral data were preprocessed with multiple scattering correction and smoothing, then developed a partial least squares regression (PLSR) model to predict the level of adulteration in minced beef. Good PLS prediction model was obtained using the full spectral range (400-900 nm) with a coefficient of determination (Rc2) of 0.9717, standard error of calibration (RMSEC) of 0.0096, coefficient of determination (Rp2) of 0.9558, and standard error from cross-validation (RMSECV) of 0.0239 by external-validation. Moreover, some important wavelengths (494, 539, 561, 585, 606, 646, 702, 770, 803nm) were selected by weighted regression coefficients(Bw) and PLS model was establish using these wavelengths. The model resulted in a coefficient of determination (Rc2) of 0.9536, RMSEC of 0.0156, coefficient of determination (Rp2) of 0.9328, and RMSECV of 0.0213. This study demonstrated using near-infrared (NIR) hyperspectral imaging technology and chemometrics have potentially to predict beef adulterated with duck.