A Rapid and Non-Destructive Approach to Identify Bone Fragments Embedded in Lean Pork Slices Based on Multispectral Imaging and Chemometrics

The feasibility of multispectral imaging in the range of 405-970 nm was assessed as a rapid and non-destructive technique for classifying lean pork slices (LPSs) and LPSs contained bone fragments. The region-of-interest (ROI-2) extracted by feature image at 700 nm played a key role to improve the precision of classification models. Seven key wavelengths (450, 470, 645, 660, 700, 780 and 970 nm) were identified using successive projections algorithm. Compared with the models established by the whole spectra, the spectral data extracted from ROIs-2 at the key wavelengths gave the highest classification of 100% for test set. A best performance for identifying the bone fragments embedded in LPSs was achieved by linear discriminant analysis model with misclassification rate less than 1% for training and test sets based on the key wavelengths. Results of satisfactory identification and visualization of bone fragments indicated that multispectral imaging could serve as an online approach for potential industrial applications.