Application of infrared spectra technique based on LS-support vector machines to the non-destructive measurement of fat content in milk powder

Fat is an important component in milk powder. It is very important to detect the fat content in milk powder fast and non-destructively. To achieve this purpose, near and mid-infrared (400 similar to 6666 cm(-1)) spectroscopy technique was used and least-squares support vector machine was applied to build a fat prediction model based on infrared spectra transmission value. The prediction result obtained from our model is better than that obtained from the back propagation neural networks (BP-NN) while the determination coefficient for prediction (R-p(2)) is 0.9796 and root mean square error for prediction (RM-SEP) is 0.8367. It is concluded that infrared spectroscopy technique can detect the fat content in milk powder fast and non-destructively, and the process is simple and easy to operate. Moreover, the prediction results based on the whole infrared spectra were compared with those based only on near infrared spectra or mid-infrared spectra data. The results show that the performances of the model based only on mid-infrared spectra or near infrared spectra data are a little worse than those based on the whole infrared spectra data.