Determination of Characteristic Wave Bands and Detection of Melamine in Fishmeal by Fourier Transform near Infrared Spectroscopy

In China, fishmeal, which is of high value and nutritionally rich, is often a target for adulteration. Melamine, which is often one of the adulterants in fishmeal, will indirectly or directly cause hazards to human or animal health. In this study, Fourier near infrared spectroscopy (FT-NIR) was applied to achieve a rapid and cost-efficient classification and quantification technique for the authentification of fishmeal. The best model parameters and the characteristic wave bands of melamine were explored. The calibration models were established based on a principal components analysis (PCA)-Euclidean distance and partial least squares (PLS) regression algorithm. Five informative spectral bands (6828–6796 cm⁻¹, 5022–4978 cm⁻¹, 4776–4763 cm⁻¹, 4637–4616 cm⁻¹ and 4496–4481 cm⁻¹) selected by the correlation coefficient method, were considered to be the characteristic wave bands of melamine. Savitzky-Golay (S–G) second derivative with smoothing over nine points and vector normalisation were the best pre-processing methods for a qualitative model; the 6873–6514 cm⁻¹ region was selected to establish a qualitative model according to the spectra of a melamine standard and fishmeal, and the correct classification rate for a set of test samples was 99.5%. The best parameters were S-G first derivative with smoothing over nine points and multiplicative scatter correction pre-processing over the spectral regions of 9947–7394 cm⁻¹ and 6915–5697 cm⁻¹ for a wide range model (samples with melamine concentrations of 0–15%). S–G first derivative with smoothing over nine points and vector normalisation pre-processing over the spectral ranges of 9947–5697 cm⁻¹ and 4605–4246 cm⁻¹ were selected for a narrow range model (samples with melamine concentrations of 0–5.5%). The root mean square error of cross-validation of the wide range model and the narrow range model were 0.396% and 0.25%, respectively; the coefficients of determination (R²) were 0.99 and 0.98; the root mean square error of prediction of testing set samples were 0.375% and 0.236%, respectively. The results demonstrate that FT-NIR can be used to detect and quantify melamine adulteration in fishmeal.