Researchs On Robustness Of Near Infrared Spectroscopy Calibration Model

The technology of Near-Infrared spectroscopy (NIR) analysis have the characters of high efficiency, quickly, low cost and nondestructive. The accuracy, reliability and application scope of NIR method depend on the calibration model developed by the databases. NIR calibration model have a very specificity, and it was set up according to the corresponding sample status and constituents. And the robustness of NIR calibration model decides the result accuracy, reliability and application scope. It is very important to set up a set of robust calibration models in application of NIR technology, and it is the basic work of transferring the NIR calibration models. The goal of this project is to setting up calibration model of wheat, rice, corn and soybean; Examining the factors which can effect the robustness of calibration models; Studying on the method to set up a well robust calibration models; Investigating the method of creating NIR calibration of single kernel seeds. The conclusions as follows: 1,It has a good coefficency and low SECV between NIR prediction value and true value in testing moisture, protein, fat, starch and gluten of wheat by NIR technology. The coefficency of wheat moisture reach 0.99, and the SECV is lower than 0.2. It is not a good result when predict the value of gel consistency and alkali spreading value. 2,The moisture, temperature, granular and status of the sample have great effects on the results and the shape of NIR. Irradiating processing have no eminent effect on NIR prediction value of wheat protein. 3,Preprocessing of NIR spectroscopy have no effect on setting up robust calibration; It is very well method to set up robust calibration with selecting a suitable scope of wavelength and taking global calibration. 4,The accuracy can be improved after getting the means of samples spectroscopy, which have been obtained by repeated scan and all directions test.