It is difficult to distinguish fresh from defrosted lamb using a single analytical test. In this study (open access) the authors propose a screening approach using Near Infra Red spectroscopy (set up online in a production environment) followed up by a panel of classical laboratory tests if required for further investigation including pH, colour parameters (L*, a*, b*), lipid oxidation (TBARS), cooking loss, and Warner-Bratzler shear force.
They used machine learning, feature selection and multivariate statistics to build classification models for each test. The models were trained on samples from twenty crossbred lamb carcasses from various commercial butcher shops. The animals were intentionally sourced from different regions of Bangladesh to ensure genetic and environmental diversity among the samples. A total of 400 meat samples were collected from these 20 carcasses, with five anatomical cuts, loin, round, rack, leg, and breast, taken from each carcass. All samples were immediately placed in sterile, ice-filled containers and transported to the laboratory then evenly divided into two groups: 200 for fresh condition analysis and 200 for frozen condition analysis. All samples were first stored at 4 °C for 24 h to allow proper post-mortem muscle-to-meat conversion. After the chilling period, the fresh group was analyzed immediately, while the remaining 200 samples were stored at −20 °C for 30 days to represent the frozen condition.
The researchers report that classification models could be built using the “classical” laboratory tests alone but they introduced the risk of overfitting. When an NIR classification model was added to the workflow as an initial screen this provided a more robust analytical approach.
[image from the publication]
