The Application Of Near Infrared Spectroscopy(NIR)in The Rapid Detection Of Feed

Near infrared spectroscopy refers to the technology using the information which contained in the absorption in near-infrared spectral region, to achieve the qualitative and quantitative analysis of materials. It could estimate the same type or similar types of unknown samples by establishing prediction models. This study mainly contained the determination of feed moisture, crude protein, crude fiber and crude fat content; collection of feed sample spectra; establishment and validation of NIR models.In this research, a total of190feed samples were collected and prepared, using chemical analysis methods to detect the moisture, crude protein, crude fibre and crude fat content. As results, the content ranged from9.72%～14.65%for moisture,14.90%～21.20%for crude protein,1.91%～4.66%for crude fibre and2.62%-4.47%for crude fat.The acquisition conditions of near infrared spectroscopy were optimized, including the scanning number and the thickness of the sample loading. Five feed samples were selected to get the optimal parameters. Results showed when under the condition of32scans and loading samples with4mm, the near infrared spectrum of the sample was the most stable and reproducible best.The quantitative analysis models of each component content in feeds were established by chemometric methods. The spectra of feed samples were collected by the instruments of FPI and Buchi. And the multivariate calibration methods used in this research included Partial Least Squares (PLS), Artificial Neural Networks (ANN) and Principal Component Analysis (PC A). The coefficient of determination (R2) of the analysis models by FPI NIR spectrometer were0.911(moisture),0.867(crude protein),0.945(crude fibre) and0.917(crude fat). The standard error of calibration (SEC%) were0.307,0.504,0.223, and0.171. The R2and SEC (%) of Buchi’s were0.9241and0.2116(moisture),0.9069and0.4008(crude protein),0.9698and0.1131(crude fibre),0.9009and0.1301(crude fat). The results showed that the FPI and Buchi feed models achieved ideal prediction effects, which had good stability and data reproducibility. The models were validated by30～50external feed samples. Results showed that the absolute errors of the prediction results and chemical values were less than0.5%without outliers, and the T-test results indicated that there was no significant difference between the two methods.The qualitative discrimination of the three different feed samples was studied, which tried to classify the samples by PCA.Conclusion: The calibration models to predict the moisture, crude protein, crude fibre and crude fat content in feed could meet the needs of quality dection. The results show that the detection error meet the provisions of national standards, which indicate the NIR models could be well applied on chemical contents prediction.