Effect Of NIR Instrument Energy Level On Model Predictive Power And Application Of OSC

The application of near-infrared (NIR) spectroscopy for estimating the various properties in samples has become widespread with the use of multivariate calibration. With NIR spectra, the analytical information is contained in small spectral variations and usually dominated by the features such as light scattering, background noise and baseline drift. NIR spectra are also affected by many factors frequently, such as material shape, size, color, and instrument status. The instrument's energy is a factor that will affect the spectra. In the first part of this thesis, the effect of NIR instrument's energy level on the model predictive power was studied with maize sample. 53 maize standard samples diffuse reflectance spectra were collected from 4000cm-1 ~10000cm-1 at 8 cm-1 resolution on Perkin-Elmer Spectrum One NTS near-infrared instrument at different energy level. 3 samples were scanned 10 times repeatedly at 100%, 76% and 34% energy level for energy variance analysis. Results show that relative standard deviation (RSD) of prediction value will become larger from 2.5% to 4.72% with energy decreasing form 100% to 18%. It is demonstrated that energy will not significantly affect predictive power by analysis of variance, because 3 samples F-value is 1.62, 3.02 and 2.23 that all less than critical value F0.05=3.35. At the same time, it is suggested that how to load samples is still an important issue in NIR diffuse reflectance analysis.Orthogonal signal correction (OSC) is provided a novel spectral pre-processing method in recent years, which is based on the orthogonal projection. This pre-processing method not only removes noise from the spectrum, but also filters the irrelevant information from response matrix. In the second part of this thesis, the nicotine and total-sugar content of tobacco andBenzene/Methylbenzene/Cyclohexane/CCl4 solution system was corrected by OSC combined with PLS. It is shown that regression models are fewer latent variables and more stable by using OSC method. The number of latent variables of nicotine model is reduced from 7 to 3; and the number of latent variables total-sugar is reduced from 6 to 3. At the same time, we also found that OSC is more excellent when it is applied on complex powder system than simple system.