Objective determination of marbling levels in raw bovine meat : using hyperspectral imaging

Abstract: Beef is a highly demanded food and plays an important role in the everyday diet for many people. The most common methods for measuring meat quality attributes are pH, colour, water-holding capacity, intramuscular fat, flavour and tenderness, are more or less performed by subjective methods and various types of chemical methods, which all are time consuming, invasive and de-structive. Furthermore, results from subjective measurements may be incon-sistent due to biased judgement. Different imaging-based techniques have re-cently been tested in order to rapidly and non-destructively measure and rank quality traits in beef cuts with consistent and reliable results. In this study imaging techniques (digital imaging and hyperspectral imaging) were used to objectively predict the average fat content and the distribution of fat on the surface of cuts of fresh beef samples. One part of this project digital (RGB) images from 61 beef samples were provided (from an earlier completed scientific study and analysed. Separately from the imaging analysis, it was evaluated how different traits such as age and breed affected the level of marbling. For each image there was back-ground information with traits (carcass characteristics and slaughter data), that was analysed using a principal component analysis (PCA). Both breed and age affected the marbling grade, but the marbling was not affected by carcass characteristics. In the other part of the project, a set of 120 commercially provided beef sam-ples were analysed using a hyperspectral camera. In order to acquire the hy-perspectral images, a hyperspectral imaging system with a spectral range of 900-2500 nm was used. Images from both datasets were analysed to obtain objectively measured fat content of all image samples. The predicted fat contents were then compared with marbling grades, evaluated by an experienced grader. The results from the image analysis indicate that both imaging techniques can be used to predict the amount and distribution of fat in beef samples, however in terms of overall accuracy, the hyperspectral imaging system was more pre-cise. The comparison between measured fat content and marbling grading in-dicate that there is a correlation between higher fat content and higher mar bling grade. However, there is a low correlation between the visually meas-ured (marbling grading) and objectively measured (RGB and HS imaging) fat content, for both the RGB dataset and the hyperspectral dataset.