Quantification of the intramuscular fat content in pork using hyperspectral imaging

INTRODUCTION
Intramuscular fat content (IMF), the fat within the lean muscle tissue, is correlated with the eating quality of meat. Once it exceeds an amount of about 2 % it becomes visible as marbling. The degree of marbling is an important visual criterion of meat quality. In the Swiss pig breeding scheme, animals are selected also according to their breeding value for IMF.
For this purpose, a quantifiable measurement of IMF is obtained by homogenizing a defined piece of meat, and analysing a sample using near infrared spectroscopy. However, this method still is laborious and does not provide any information on the coarseness or spatial distribution of the intramuscular fat. It therefore is also not applicable in an industrial routine process to sort meat cuts according to their IMF or marbling.
Hyperspectral imaging is an analysis technique that combines spectroscopy and imaging, thus providing both chemical and spatial information at the same time. Through the use of a hyperspectral camera, an image can be collected of a sample in a few seconds, where each pixel in the collected image contains a spectrum representing its light absorbing properties. By applying multivariate data analysis methods like PLS, a calibration model can be obtained from the hyperspectral image that can be used to measure chemical content, for example fat %, in each pixel of the collected image.
EXPERIMENTAL
For this study a SWIR (Shortwave Infrared) hyperspectral camera (Specim) was used, spectral range 1000 – 2500 nm, 320 spatial pixels and 256 spectral pixels. A field of view of 200 mm was used giving an image with a pixel size of 0.625 mm/pixel. The images were analysed and a calibration model was developed using a pre-treatment of auto scaling and SNV (Standard Normal Variate) transformation followed by PLS. All data analysis was done using the Evince Image software (UmBio). The training set consisted of 30 samples of pork meat with an IMF value ranging from 0.5 to 5% (measured using an existing reference method). To validate the calibration, a test set of 16 samples were predicted using the calibration, and the predicted result were compared to the actual values.
RESULTS AND DISCUSSION
A calibration model for IMF% with an R2=0.92, Q2=0.91 and RMSE=0.21% was obtained, indicating that hyperspectral imaging can be used as a fast and accurate method to determine the average content of intramuscular fat in a sample of pork. It should also be possible to apply image analysis algorithms on the recorded hyperspectral data to determine the coarseness and the spatial distribution of the IMF in each sample, which then, combined with the determined IMF%, would provide an objective, fast and easy method for grading the marbling in meat.