Transfer of NIR Prediction Models for Main Chemical Components in Tobacco

Abstract

The transfer of NIR prediction models for main chemical components in tobacco between 2 NIR devices of the same type was investigated. The results indicated that: 1) the NIR models for total alkaloids, total sugar, reducing sugar, total nitrogen, potassium, total volatile acids, total volatile bases, petroleum ether extract and its neutral proportion, polyphenols, sulfate contents, and pH value of tobacco could be directly transferred, while those for chlorine, starch, cellulose, ash, and water soluble ash alkalinity could not; 2) standard errors of prediction of models modified by slope/intercept algorithm and multi-sample stan-dardization for chlorine, starch, cellulose, ash and water soluble ash alkalinity were 0.061, 0.523, 1.250, 1.195, 0.206 and 0.062, 0.512, 1.129, 1.261, 0.205, respectively. The prediction precision of modified models improved by 75.20%, 55.37%, 36.19%, 37.37%, 71.23% and 74.80%, 57.48%, 42.37%, 33.91%, 71.37%, respectively, which was comparable with that of master instrument, and there was no significant difference between the prediction precision of the 2 modified models.