Uninformative Variable Elimination Combination With Direct Orthogonalization In Near-infrared Spectral Analysis

Near Infrared detection technology has many advantages such as quickanalysis speed, large amount of spectral information, no sample destroy, easy onlineanalysis, low-cost analysis, no pollution, and real-time monitoring, etc,, so that itreceived considerable attention. But the characteristic of Near Infrared such as widespectral region, overlapping peaks, large search space, and spectralacquisition often subjects to environmental noise and interference of othercomponents. These problems, in recent year, are the bottleneck that always restrictsNIR further development.Using near infrared detection technology，dairy and other organic solution asthe main research object has been detected and researched; Near-infrared spectraldata was obtained from the preparation of four-component organic solution and thecollection of milk; PLS model of obtained spectra was respectively established afterspectral progressing and variable selection, then the composition and content ofmaterial was determined.In accordance with the problem such as influence of external noise and mutualinterference between components in near-infrared spectroscopy, this paper adoptedDirect Orthogonalization to pre-treat spectra, which can make use of theorthogonality between non-target information and target information to make lesseven eliminate all non-target factors that impact on the spectrum. In view ofspectrum itself absorbance low, wide spectral region and spectral overlap,Uninformative Variable Elimination is used to select variable, which can eliminatethe non-test sample variable information of near-infrared spectrum, and only specificrange spectrum is purposefully collected to get spectral data, thereby overlappedpeaks and complex spectrum were simplified, detection time was shorten. At thesame time, this paper attempted to combine above methods to U-D-PLS(Uninformative Variable Elimination-Direct Orthogonalization-Partial Least Squares) to respectively progress the spectra of homemade four-component organic solutionand dairy, then establish PLS model, obtain UVE-DO-PLS(U-D-PLS) model. Finally,the model evaluation is used to assess whether this combination could make modelget a greater degree of optimization.Near infrared spectroscopy was used as detection means and combined withchemometric methods to establish the model between sample concentrations andspectral. The robustness and predictive ability of model was improved through thecombination of different algorithms. Dairy and other organic solution are also theimportant components of organisms, so this study not only applies to dairycomponents testing, but also can be used in other food, pharmaceutical industry,agricultural testing and so on.