Criterions of meat marbling characterization by RGB-, hyperspectral and CT imaging algorithms

Abstract

Danbred pigs (n=164) of both sexes were fed with different lysin/DE (digestible energy) ratio feed. The marbling characteristics of chop and loin samples were described by optical/digital imaging methods, with evaluation of experts and chemical analysis including fat, dry matter and protein content. For optical estimation of marbling and fat content, the criterions of objective image processing algorithm were set as: result must correlate with scores of experts and analytical measurement, must be invariant from rotation, resolution, illumination, should result similar scores for chuck and chop tissues of same animal, finally the score should be mostly constant on different cross-sections of same tissue. Scattering, RGB, NIR hyperspectral and CT images were taken for each sample. On RGB images, the algorithm being invariant from rotation, resolution and illumination, resulted scores which correlated well with expert and analytical values. The scores of different tissues of same animal were moreover in relation. On the base of hyperspectral measurements first the significant wavelengths of lean, fat and connective tissues were determined. In this work, the simple fat contra lean graylevel images were used for testing efficiency and invariability of image processing algorithm. On cross-sections of CT, although the pattern is significantly changing along fibre direction, the score of algorithm was broadly constant. The criterions of correlation, invariability and stability were fulfilled for the algorithm.