Feasibility Study of Discriminating and Quantifying Low Levels of Melamine Contamination in Fishmeal by Fourier Transform near Infrared Spectroscopy

This study was conducted to demonstrate the feasibility of Fourier near-infrared spectroscopy (FT-NIRS) to detect and quantifying low concentrations (3.00–1056.80mg kg-1) of melamine contamination in fishmeal and to choose a better chemo metric method by comparing the results of the models based on different chemo metric methods. The qualitative calibration models were established based on PCA-Euclidean Distance and least squares-support vector machine (LS-SVM) respectively and the quantitative calibration models were established based on partial least squares (PLS) regression algorithm and least squares-support vector machine (LS-SVM) respectively. Savitzky-Golay second derivative with smoothing over five points and vector normalization were the best pre-processing methods. A qualitative model, established based on this pre-processing method, was capable of identifying the testing set samples with melamine concentrations higher than 136mg kg-1, with a 100% correct classification rate. Further, the qualitative models based on PCA-Euclidean distance, S-G first derivative with smoothing over nine points and vector normalization pre-processing methods and the frequency ranges of 9099-8246 cm-1 and 7398-6545cm-1 were the best parameters selected by the optimizing process. Quantitative models based on these parameters accurately predicted the samples with melamine concentration of higher than 208mg kg-1, with the mean relative forecasting deviation less than 5%. The model based on LS-SVM was obviously not better than that based on PLS. The results show that FT-NIR can be used to detect and quantify low concentrations of melamine contamination in fishmeal.