Chemometric processing of visible and near infrared reflectance spectra for species identification in selected raw homogenised meats

Visible and near infrared reflectance spectra (400-2498 nm) of 230 homogeni sed meat samples (chicken, turkey, pork, beef and lamb) were collected. Cla ssification of the spectra into individual species was attempted using fact orial discriminant analysis (FDA), soft independent modelling of class anal ogy (SIMCA), K-nearest neighbour analysis and discriminant partial least sq uares (PLS) regression. A variety of wavelength ranges and data pretreatmen ts were investigated for optimum accuracy. Particular difficulty was encoun tered in distinguishing between chicken and turkey; models were, therefore, initially developed using five separate meat classes and again using four groups, with chicken and turkey being amalgamated into a single class. In a four-group classification, the best models produced between 85 and 100% co rrect identifications. Using five groups, classification rates were general ly lower. FDA and PLS discrimination generally produced the best accuracy r ates. SIMCA exhibited the poorest classification performance.