Improving the robustness of early prediction of pork ultimate pH using Raman spectroscopy on the slaughter line

Les Cahiers de l'IFIP, 9(1), 47-56.
The first set of trials to performance-test a Raman device purpose-designed for early on-line prediction of pork-meat quality (ultimate pH and drip loss) was led by the IFIP in 2017 on pigs slaughtered in France, in collaboration with the University of Bayreuth. The prototype Raman device confirmed fitness-for-purpose in real-world industry-scale slaughterhouse conditions.
This second IFIP-led project sets out to strengthen the original database developed in 2017 by selecting carcasses that present pH24 extremes. The effort to select carcasses early upstream compelled the use of a destructive biochemical method to gauge ultimate pH on the slaughter line. Mobilizing this method enabled us to apply a selection pressure so as to create a calibration population featuring dense variability, which is a key factor for improving predictive precision. This study also features a strand of work addressing on-slaughterline measurements in industry-scale conditions at high line throughput and with the shortest workable integration time. New models developed here based on the population set strengthened with extreme-value samples showed increased robustness in cross-validation and external validation, with a good error level in external validation. This fresh science raises prospects for using early-measurement Raman spectroscopy to robustly predict pH24 on hot pre-rigor carcasses.
The lack of precision for the prediction of drip loss is explained by difficulties methodizing out-sorting time. The Raman spectra acquired on-line with a 3 s capture per carcass nevertheless show that it is possible to correlate pre-rigor spectral measurement to pH24, but not yet precisely enough. Good performances can be achieved using a 2-class statistical classification model.
This study has served to develop precise and robust calibrations for early prediction of pork carcass pH24. The Raman device will nevertheless require further substantial modifications—notably to address signal integration times and signal-to-noise ratios—before it can be readied for use in industry-practice slaughter-line conditions.