Grundlagenuntersuchung zur Raman-Sensorik von Lactat für eine automatisierbare Beurteilung der Fleischqualität in der Prozesskette
Plant Cultivation, Plant Nutrition, Agricultural Technology
Final Report Abstract
The ultimate quality of the porcine meat can be estimated from the pH changes occurring during the first 24 h post mortem. These pH changes result from postmortem glycolysis and specifically from the conversion of glycogen into lactate. Variation in the rate and the extent of this conversion is responsible for the different meat characteristics and implicitly for its freshness. Thus monitoring the pH dynamics can be potentially used as an indicator for the meat quality evolution. In this investigation, the use of Raman spectroscopy in combination with chemometric evaluation was examined as a fast tool for the prediction of pH and lactate determination as well as overall meat quality. While Raman spectroscopy proved its use as a tool in chemometrics, predicting meat metabolic parameters using this spectroscopic method is a relatively new approach. It could be demonstrated, that data pre-processing prior to modeling is mandatory to remove irrelevant features in Raman spectra that will reduce the modeling performance. Using a comparative approach we also established that for the linear and non-linear regression models, spectral normalization techniques, such as standard normal variate (SNV) and multiplicative scatter correction (MSC), managed to significantly increase the performance of the chemometric models for the lactate and pH prediction. Additionally we also employed the ACO and GA metaheuristics to allow data reduction by up to 90 % without significantly impairing the quality of prediction. The obtained results suggests that a combination of the locally weighted regression and the SNV transformation is the best pre-processing procedure for this application. These pre-processed Raman spectra provides one of the most accurate and robust models with a cross-validated coefficient of determination (r2cv) of 0.97 for pH and lactate, a cross-validated root mean square error (RMSECV) of 4.5 mmol/kg for the lactate prediction and 0.06 pH-units for the pH prediction. These results demonstrate the great potential of combining Raman spectroscopy and chemometric evaluation for online meat quality control applications. Furthermore, the potential of combining Raman spectroscopy with chemometric evaluation was examined into predicting the meat quality. As indicator for meat quality the predictability of the actual pH value as well as the pH evolution within 24 h post-mortem, which is critical for the meat end-quality, out of Raman spectra are used. The results obtained using ACO metaheuristic approach show that the early meat pH measured within 45 min post-mortem (pH45) can be predicted with relatively similar accuracy out of both, Raman spectra recorded at the initial measurement time as well as from the spectra recorded 24 h later. Interestingly the vice-versa crossprediction approach works in a similar manner as the meat late pH measured after 24 h (pH24) can be predicted out of the spectra recorded 24 h post-mortem as well as from the initial spectra recorded within 45 min. Moreover, the prediction of the meat pH decline within 24h post-mortem has also proved feasible out of the initial spectral recording contained in the pH 45 dataset. Since the used metaheuristics are often regarded as blind-search methods, which will select relevant and generic wavelengths alike, their initial search space has been altered by adding a priori knowledge and thereby by decreasing the probability of selection for the non-relevant wavelengths. Additionally have been conducted Monte Carlo repetitions for the ACO and GA metaheuristic runs, followed by a statistic of its selection outcome, which seem to improve the stochastic solutions usually delivered by metaheuristics. This translates into the fact that the metaheuristics selects the same spectral regions disregarding the number of runs it performs. Although some positive effects were observed in regard to accuracy of the weighted models, they show a declined coefficient of determination in comparison with the un-weighted models which suggested that the list of the pH-relevant wavelengths in the Raman spectra may be still incomplete. These results won’t disregard the necessity of the weighting, but merely acknowledge the fact that there is still much unknown information concerning the pH-related fingerprint localization in Raman spectra of porcine meat. In conclusion, the obtained results of the prediction/cross-prediction of the pH evolution out of different time frame Raman recordings using the weighted metaheuristics suggests that this information which can be quantified has different fingerprint origins that can be extracted from of any of the Raman spectra taken within that time frame. In the meat production industry, where the classical quality related-pH measurement is rarely performed due to the fact that it is slow and promotes contamination, such an alternative way to measure and predict meat quality evolution could prove to be real advantageous.
Publications
- Raman-Sensorik zur automatisierbaren Beurteilung der Fleischqualität, Fleischwirtschaft, 93 (2013), pp.170-174
Scheier, R. M. Nache; N. Agarkov, B. Hitzmann, H. Schmidt
- Meat Quality Prediction Using Raman Spectroscopy and Chemometrics, Journal of Cheminformatics 6. Suppl 1 (2014), p. 21
Nache, M.; R. Scheier, H. Schmidt, B. Hitzmann
(See online at https://doi.org/10.1186/1758-2946-6-S1-P21) - Non-invasive lactate- and pH-monitoring in porcine meat using Raman spectroscopy and chemometrics, Chemometr. Intell. Lab. (2015) 142, pp. 197-205
Nache, M.; R. Scheier, H. Schmidt, B. Hitzmann
(See online at https://doi.org/10.1016/j.chemolab.2015.02.002) - Non-invasive lactate- and pH-monitoring in porcine meat using Raman spectroscopy and chemometrics. Chemometrics and Intelligent Laboratory Systems, Volume 142, 15 March 2015, Pages 197-205
Nache, M.; R. Scheier, H. Schmidt, B. Hitzmann
(See online at https://doi.org/10.1016/j.chemolab.2015.02.002)
