Application Of Pre-Processing Of Nirs Modeling Data

Wang Zhihong,Lin Jun

2006-01-01

Abstract:Based on the difference of acting-region and aim and effect, the two classes of pre-processing methods for NIR spectrum data have been concluded. They are the spectrum and matrix pretreatment. Their characteristic and application are the studied by experiments. The spectrum pretreatment to the spectrum matrix can be used to eliminate or decrease the infections of the factors that are not relation to the sample component or structure. The matrix pretreatment to the spectrum and/or consistence matrix can be used to increase the difference between the modeling data. Their application can be unite, and their order is spectrum. matrix. When they were used separately. and the approaches are appropriate. such as standardization or normalization spectrum pretreatment and centralization matrix pretreatment. the model accuracy of the corn data is near or over 95%. When they were used unitedly, the effect of each united approaches is different. So in order to heighten the accuracy of the model, the choice of data pretreatment approaches should be based on the mutual benefit principles.