Study Of Detecting And Quantifying Melamine Adulterated In Fish Meal By Fourier Transform Near Infrared Spectroscopy

The study was aimed at exploring the feasibility and methods of detecting and quantifying melamine adulterated in fish meal by Fourier transform near infrared spectroscopy (FT-NIR). So in the first test, a total of 126 commercial fed fish meal samples were collected, and 276 adulterated samples were prepared by adding different amount of melamine standard to the fed fish meal samples, melamine concentration of 0.1%-15.0%. The spectral data was collected by FT-NIR. A qualitative model was established based on PCA-Euclidean Distance; quantitive model was established by partial least squares (PLS) regression algorithm. In the second test, to exploring the detection ranges of quantifying low concentration melamine contaminated in fish meal further, at the same time, also to comparing the effects of the different chemometrics methods on NIR model. Selecting 55 commercial fed fish meal samples from above,60 contaminated fish meal were prepared by adding different amount of melamine standard to these fed fish meal samples, melamine concentration of 3.0 mg/kg-1056.8 mg/kg. The spectral datas were collected by NIR. Qualitative models were established based on PCA-Euclidean Distance and LS-SVM, respectively; quantitive models were established by PLS regression algorithm and LS-SVM respectively.The results showed as follows:Five informative wave bands (6828-6796,5022-4978, 4776-4763,4637-4616 and 4496-4481 cm-1) selected by correlation coefficient method, are considered to be the characteristic wavenumbers of melamine, and 0.4 was the threshold; When the melamine concentration of 0.1%～15.0%,6873-6514 cm-1 were the best spectral range of qualitative analysis model, and the correct classification rate of the qualitative analysis model to discriminating the testing set samples was 99.5%.9947-7394 cm-1, 6915-5697 cm-1 were the optimum spectral regions of wide range quantitive model, according to the spectral optimization method of successively removing low-information spectral regions. The R2, RMSECV and RMSEP of testing set by the model were 0.99, 0.40% and 0.38%, respectively; When the melamine concentration of 0.1%～5.5%, 9947-5697,4605-4246 cm-1 were the optimum spectral regions of narrow range quantitive model according to the sepctral optimization method as mentioned above. R2, RMSECV and RMSEP of testing set by the model were 0.98,0.25% and 0.24%, respectively; When the melamine concentration of 3.0 mg/kg-1056.8 mg/kg, the correct classification rate of the qualitative analysis model established on PCA-Euclidean Distance at 6873-6514 cm-1 to detecting testing set samples was 70.4%, and the correct classification rate of samples with melamine concentration of 136.4 mg/kg or higher was 100%. Qualitative analysis model established on LS-SVM at 6873-6514 cm-1, the correct classification rate of testing set samples was 88.9%, and the correct classification rate of samples with melamine concentration of 136.4 mg/kg or higher was 100%. Qualitative analysis model established on LS-SVM at 5300-4900 cm-1 or full spectral range, the correct classification rate of testing set samples was 81.5%, and the correct classification rate of samples melamine concentration of 136.4 mg/kg or higher was 100%; When the melamine concentration of 3.0 mg/kg-1056.8 mg/kg,9099-8246 cm-1,7398-6545 cm-1 were the optimum spectral regions of low-dose quantitive model according to the sepctral optimization method mentioned above. The R2, RMSECV and RMSEP of testing set were 0.93,81.7 mg/kg and 66.5 mg/kg, respectively. The low-dose quantitive model based on LS-SVM, the R2, RMSECV and RMSEP of testing set were 1,71.0 mg/kg and 118.9 mg/kg, respectively. The relative predicted deviations were small of the testing set samples with melamine concentration of 224.5 mg/kg or higher by the low-dose quantitive models based on PLS or LS-SVM, and the RMSEP were 37.7 mg/kg and 37.82 mg/kg, with higher accurate prediction; NIR can be used to predict the crude protein and moisture of fish meal, the R2, RMSECV and RMSEP were 0.92 and 0.91,0.40% and 1.39%,0.67% and 2.99%, respectively. The R2, RMSECV and RMSEP of predicting crude ash of fish meal model was 0.85,1.99% and 1.66%, respectively. The coefficient of determination (R2) of lysine, aspartate, threonine, glumatic acid, glycine, alanine, valine acid, isoleucine, leucine, tyrosine, phenylalanine, arginine, proline, total amino acid of fish meal were all higher than 0.9, predicted relative standard deviation were less than 10%; the R2 of NIR medel of methionine was 0.85, predicted relative standard deviation less than 10%; the R2 of NIR medel of histidine was 0.92, but the predicted relative standard deviation higher than 10%; R2 of serine and cystine NIR models were 0.62 and 0.47, and the predicted relative standard deviation higher than 10% both。In conclusion:the characteristic wave bands of melamine were selected by correlation coefficient method,0.4 was the threshold; when the melamine concentration of 0.1%～15.0%, NIR can quickly and exactly detect whether fish meal was adulterated with melamine or not and predicting melamine content, with a higher accurate prediction of samples with mealmine concentration of 0.1%-5.5 by the narrow range model; FT-NIR can detect whether fish meal was contaminated with low-dose melamine of 136.4 mg/kg or higher by qualitive analysis; the qualitative model established by LS-SVM was better than the model by PCA-Euclidean distance; the quantitive low-dose model (based on PLS) can be used to predict the amont of melamine contaminated in fish meal at the concentration of 208.3mg/kg or higher, the RMSECV was 81.7 mg/kg, and the RMSEP of model to the testing samples with melamine concentration higher than 224.5 mg/kg was 37.7 mg/kg. The quantitive low-dose model established by LS-SVM was not better than by PLS obviously; NIR can predict the moisture, crude protein and a part of amino acid of fish meal.