34377

Discrimination and prediction of chemical composition of the fish species from Amazon coast using a long wave near infrared (LW-NIR) portable instrument

Favoritar este trabalho

NIR vibrational spectroscopy has many applications in the food industry, particularly in quality control, discrimination and determination of chemical compositions of various foodstuffs because of the speed and ease of application in routine analysis when compared to the traditional physical and chemical methods. Thus, the aim of this study was to evaluate the efficiency of NIR to discriminate five species from the Amazon coast (Genyatremus luteus, Lutjanus purpureus, Macrodon ancylodon, Cynoscion acoupa, Micropogonias furnieri) and predict major chemical components The spectral measurements were made using a portable spectrophotometer which uses a long wavelength band located within the NIR region (1600-2400 nm). Fish fillets of these species were analysed in the intact, ground and freeze-dried forms. The spectra were pre-processed by derivatives using the Savitzky-Golay algorithm for smoothing the spectral noise, both for the use of prediction and discrimination models. The models obtained by principal component (PCA) and Soft independent modelling by class analogy (SIMCA) were applied for discrimination of species. For the prediction of major components, it was used the Principal Component Regression (PCR) and Partial Least Square regression (PLS). With PCA, it was observed the formation of groups related to the species G. luteus and L. purpureus in intact and lyophilized samples, while with SIMCA, the formation of groups of the five species was more significant, confirmed by the distance between the groups in the range from 4.16 to 13.31 and 2.90 to 57.05, for lyophilized and intact samples, respectively. The PLS regression model showed a higher value for the coefficient of determination (r2) of cross validation than PCR. For the calibration coefficient of determination (R2) with lyophilized samples, it also showed better results. For all three major components, it was evaluated the values of the residual predictive deviation (RDP) and root mean square of prediction for the cross-validation set (RMSEPCV). The major components were presented: moisture (r2: 0.58, RMSEPcv: 1.11 and RDP: 1.40), lipids (r2: 0.96, RMSEPcv: 1.09 and RDP: 4.82), and total protein (r2: 0.86, RMSEPcv: 1.96 and RDP: 2.60). A good correlation between data obtained by AOAC methods and predicted by NIR was obtained.