| In this study, domestic Near-infrared Analyzer (1000-1800nm) combined with the chemometric methods were used to develop near infrared reflectance spectroscopy (NIRS) calibration models for predicting chemical composition and some physical parameters of beef, and comparative analysis were conducted to evaluate the prediction accuracy of models.Experiment1, the objective of this experiment is to develop NIRS models for predicting the chemical composition of beef from commercial cuts. A total of442minced beef samples from various commercial cuts of two cattle breeds were used for the NIRS modeling process and randomly divided into two subsets:a calibration set and an independent prediction set (75%vs.25%). Reflectance spectra (1000-1800nm) were collected from both subsets of samples, and calibration models were built using partial least squares regression (PLSR) on the calibration set of samples. Different mathematical pre-treatments were tested and mean centering (MC) combined with1st derivative pre-processing gave the best calibration models on all the beef chemical compositions. According to the selected calibration equations, both the coefficient of determination in calibration (R2c) and the coefficient of determination in prediction (R2p) were over0.88for all chemical compositions. The ratio performance deviation (RPD) was between2.0and3.0for fat, protein and moisture. The results of the present study indicate NIRS could be used for rapid prediction of some chemical composition in the beef, but the predictive ability still need further improvements before practical application. Experiment2, the objective of this experiment is to develop NIRS models for predicting some physical parameters of beef from commercial cuts. Spectra collection and modeling approaches are the same with experiment1. Different mathematical pre-treatments were tested and MC or multiplicative scatter correction (MSC) combined with1st derivative pre-processing gave the best calibration models on all the beef physical traits. According to the selected calibration equations, the model for pH prediction was reluctant to accept (R2C=0.723, R2P=0.730, SEP=0.202, RPD=1.53), while the model for Warner-Bratzler shear force (WBSF) in the calibration was acceptable (R2C=0.793), but the predictive ability on independent validation set was poor (R2P=0.661, SEP =1.324, RPD=1.31). High SEP combined with low R2P and RDP indicated the prediction abilities of NIRS for cooking loss and color parameters (L*, a*and b*) on minced beef samples were unsatisfactory (R2P<0.70, RPD<1.3). Although NIRS has the potential for rapid prediction of some quality parameters in the beef samples, further improvements are required before practical use can be considered.Experiment3, the objective of this study was to develop robust calibrations of NIRS to maximize the predictability of the main chemical composition of beef. A total of182minced beef samples from various commercial cuts of four cattle breeds and some artificially mixed samples (with high proportion of fat tissue) were used for the NIRS modeling process and randomly divided into calibration set (n=140) and independent validation set (n=42). Reference values of chemical composition had extremely wide ranges (fat=0.20-86.45%, protein=1.98-23.41%, moisture=12.85-79.25%). Spectra collection, laboratory analysis and modeling approaches are the same with experiment1. According to the selected calibration equations, both R2C and R2P were over0.98for all chemical compositions. The RPD was17.37,5.12and10.43for fat, protein and moisture, respectively. The results of the present study indicate the outstanding ability of NIRS to predict chemical composition in beef from different cattle breeds, which was probably due to the wide ranges of reference data and the homogeneity in minced beef samples. To our knowledge, performances of the calibration equations have never been so high to offer an alternative to analytical methods of the chemical composition. |
PDF Full Text Request