Challenges in Model Development for Meat Composition Using Multipoint NIR Spectroscopy from At-Line to In-Line Monitoring.

This study evaluates the efficiency of multipoint near-infrared spectroscopy (NIRS) to predict the fat and moisture content of minced beef samples both in at-line and on-line modes. Additionally, it aims at identifying the obstacles that can be encountered in the path of performing in-line monitoring. Near-infrared (NIR) reflectance spectra of minced beef samples were collected using an NIR spectrophotometer, employing a Fabry-Perot interferometer. Partial least squares regression (PLSR) models based on reference values from proximate analysis yielded calibration coefficients of determination (Rc2) of 0.96 for both fat and moisture. For an independent batch of samples, fat was estimated with a prediction coefficient of determination (Rp2) of 0.87 and 0.82 for the samples in at-line and on-line modes, respectively. All the models were found to have good prediction accuracy; however, a higher bias was observed for predictions under on-line mode. Overall results from this study illustrate that multipoint NIR systems combined with multivariate analysis has potential as a process analytical technology (PAT) tool for monitoring process parameters such as fat and moisture in the meat industry, providing real-time spectral and spatial information.