Ability of near Infrared Spectroscopy to Predict Pork Technological Traits

Abstract

The predictive ability of near infrared spectroscopy was studied for some fresh meat quality traits: pH24, colour parameters (L*, a* and b*) and drip loss (measured after 24 and 48 hours). The material used in this study involved 296 pig longissimus dorsi muscle samples originating from various hybrids of different carcass weight, two abattoirs and five series of slaughter. Samples were scanned both intact and minced over the wavelength range 400–2500 nm using an NIRSystems model 6500 spectrophotometer. Modified partial least squares (PLS) was used to develop models and to obtain calibration statistics: coefficient of determination of calibration and cross-validation ( R2 CV) and standard error of calibration and cross-validation ( SECV). The predictive ability of near infrared (NIR) spectroscopy was additionally tested on an independent set of samples and prediction statistics calculated. Best calibrations were obtained on intact meat samples using the spectral range 400–1100 nm. For these models, the R2 CV values were in the range of 0.54–0.79 and SECV values were 0.10 for pH24, 2.4 for L*, 0.8 for a*, 0.9 for b*, 1.6% for drip24 and 2.1% for drip48. Within-sample standard deviation (repeatability) of reference measurements was 0.05 for pH24, 3.1 for L*, 1.0 for a*, 0.9 for b*, 1.1% for drip24 and 1.3% for drip48, indicating limitations in NIR spectroscopy's predictive ability due to the precision of reference methods used for calibration. Prediction errors obtained on the validation set (0.08 for pH24, 2.0 for L*, 0.8 for a*, 0.7 for b*, 1.4% for drip24 and 1.7% for drip48) were in agreement with cross-validation results. Similar model performance was found for driploss prediction (PLS regression) based on spectral information or on a combination of pH24, L*, a*, b*.