Interpretive Near Infrared Spectroscopy Technology And Model Establishment

Near infrared spectrum signal spectrum peaks overlap was serious.The traditional chemical metrology method didn't consider the physical and chemical meaning of spectrum.Any link's change could influence the results of the near infrared models,so this kind of analysis results weren't entirely convincing.It has long been the research hotspot and difficulty in the near infrared field.Based on the near infrared spectrum,the research work mainly included two parts,respectively,the interpretive near infrared spectroscopy technology,and near infrared spectroscopy quantitative and qualitative analysis model,the purpose was to find the spectrum information which was directly related to the target analysis composition,and build more explanatory,convincing near infrared model.The main contents were as follows:(1)A new interpretive spectrology method was proposed,which was an improved ICA algorithm and was named as independent characteristic projection.Moisture,corn oil,zein and corn starch were the four main components of corn.five mixed imitation corn samples were prepared in a certain proportion.The feasibility of using independent characteristic projection for interpretive spectrology was verified by analyzing their spectral characteristics.Corn spectra were decomposed respectively based on four bases,three bases,two bases or single bases,by independent characteristic projection,so as for the interpretive spectrology.(2)The quantitative analysis models of the four corn components were established based on the decomposed spectra,and were compared with the model of the original corn spectrum.The results showed that the independent characteristic projection played a positive role in the interpretable near infrared model.The influence of two factors,which were the number and the combination of based spectra,on the model was analyzed and discussed,and the combination priority of based spectra was figured out.(3)The effect of four pretreatment methods were compared.That the CARS algorithm could quickly extract effective information from spectrum was verified.The results showed that the first derivative was the best,whose model recognition rate is 86.11%,and it increases to 97.22% after using the CARS algorithm.