Quantification of Intramuscular Fat Content in Beef by Combining Autofluorescence Spectra and Autofluorescence Images

Abstract

Autofluorescence spectra, autofluorescence images, and near-infrared (NIR) reflectance spectra were tested to determine intramuscular fat and connective tissue in meat slices (longissimus dorsi) from 45 Norwegian Red cattle. Excitation wavelength 332 nm was used to generate fluorescence that was spectroscopically measured and imaged over a total area of 33 and 21 cm2, respectively, for each sample. Five types of image features were extracted: the angle measurement technique, the autocovariance spectrum, measures derived from the gray-level co-occurrence matrix, singular value decomposition, and percentage of bright areas in the images. The fat content was 0.5–10.5% (w/wet w) and connective tissue 0.42–0.92% (w/wet w). Partial least-squares regression resulted in the following correlations ( R) and root mean square errors of cross-validation (RMSECV) for fat: autofluorescence spectroscopy (0.84, 1.10%), NIR (0.87, 1.01%), and the best image feature (0.78, 1.30%). The best result was obtained by combining autofluorescence spectra with image features (0.92, 0.80%). Connective tissue was poorly predicted ( R = 0.41) because of narrow range. The study demonstrates that the combination of spectral data and structural information from images can result in better and more robust prediction models.