Quantitative and non-destructive evaluation of ground beef based on multi-spectral imaging

The quality of meat-based products is usually tested by subjective and analytical methods. There are laboratory tests which can accurately estimate the content of a sample. Yet, they imply the sample destruction. In addition, they are time consuming and not suitable for industrial applications. Spectral unmixing is wide popular in remote sensing and biomedical applications for a quantitative analysis of an image. In this study, we apply an optical characterization of a ground-beef sample by blind linear unmixing. We prepare samples of ground beef with fixed fat/protein content. The samples are employed to evaluate the characterization provided by linear unmixing of multi-spectral data. We use an eight-band multi-spectral camera and halogen lamps as illumination source. A constrained quadratic optimization algorithm is employed to estimate end members and their abundances in the sample. The linear unmixing was applied to estimate four end-members and their abundances in the ground beef samples. These abundances match the visual characteristics of the sample such as positions with high concentration of fat.