Species identification in selected raw homogenized meats by reflectance spectroscopy in the mid-infrared, near-infrared, and visible ranges

Mid-infrared (2000-800 cm(-1)), near-infrared (750-2498 nm), and visible (400-750 nm) reflectance spectra of 230 homogenized meat samples (chicken, turkey, pork, beef, and lamb) were collected. Species identification was attempted by using factorial discriminant analysis (FDA), soft independent modeling of class analogy (SIMCA), K-nearest neighbor (KNN), analysis and discriminant partial least-squares (PLS) regression. A variety of wavelength ranges was investigated for optimum accuracy. Particular difficulty was encountered in distinguishing between chicken and turkey; models were therefore developed with the use of five separate meat classes and again with the use of four, with chicken and turkey samples being treated as one group. Discriminant PLS, FDA, and KNN models provided similar levels of accuracy in this application. Correct classification rates in excess of 90% were achieved in all cases.