In this paper, a NIR Spectral System based on Acoustic Optical Tunable Filter (AOTF) was developed for measuring milk constituents. Factors affecting the repeatability of measuring spectra of the system were analyzed by using the transmitted spectra of Ultra-High Temperature (UHT) sterilized milk. Orthogonal experimental design method was employed to determine the significance level of these factors and the optimal condition for measuring spectra of the system. It is concluded that the crucial factor affecting the spectra repeatability is the ambient temperature. Further temperature testing results show that, when the environmental condition is basically invariant, the temperature of the AOTF crystal is basically constant and the drifting amplitude of wave number is within 0.5cm-1, when the ambient temperature varies, the temperature of AOTF crystal changes correspondingly. The drifting of wave number is about 1cm-1 when the temperature of AOTF crystal changes 1°C. It is found that the wave number drifting increases with decreasing temperature of AOTF crystal. These testing results are of important significance to further improve the measuring accuracy of the system.
In this paper, methods of adopting NIR spectra to measure milk constituents and Direct Orthogonalization (DO) pre-processing are studied. Based on the spectra of representative solutions, the DO method was employed to filter the signal noise that is irrelevant to the concentration being measured. The predictions of the Partial Least Squares (PLS) Regression models with and without the DO pre-processing are evaluated. During the experiments, the transmitted spectra of a water solution with glucose and NaCl mixture and the scattered reflection spectra of milk are acquired respectively in the near infrared region of 1000~1700nm using a homemade NIR Spectral System for measuring milk constituents. With the DO pre-processing of the spectroscopic data, the number of the optimal Principal Component (PC) of the PLS model is reduced. It should be noted that the sum of this PC number and the PC number corresponding to the DO preprocessing is equal to the optimal PC number of the PLS model without DO pre-processing. After the DO pre-processing, the Root Mean Square Error of Cross Validation (RMSECV) of PLS model is slightly reduced whereas the Root Mean Square Error of Prediction (RMSEP) is reduced significantly.
This paper presents a near-infrared spectroscopy method for measuring milk constituents based on AOTF technique. Some key problems encountered during study of spectrum stability, such as lower stability, parallel moving of the spectrum, were analyzed from the point of view of the system structure and the AOTF driver, The reliability designs, such as heat design, EMC design, and software allowance design, were performed. The capacities against interference and spectrum stability ofthe system have been improved accordingly, the coefficient of variation ofrepeatability within short time-scale has reached to 0.0002, and stability within long time-scale has reached to 0.001.