Model Optimization And Stability Of Near Infrared Diffuse Reflection Spectroscopy Analysis Of White Granulated Sugar

Using near infrared diffuse reflection spectroscopy technology and chemometrics method establish color value of white granulated sugar model of quantitative analysis.The method is based on the best validation of all the samples of single-point model,which can best divide calibration sample set and validation sample set.And based on this method, using all kinds of new chemometrics techniques systematic establish the calibration and validation model of near infrared diffuse reflection spectroscopy Analysis of color value of white granulated sugar.In the the goal of the effect of predict,debugging the various model parameters,after a large number of computer numerical experiments carrying out and in large-scale screening optimizing models,it can be determined the effect of high validation characteristics wavelength-bands and wavelength-points combinations.And through all kinds of different devide of calibration sample set and validation sample set to test the stability of models.For the Near infrared diffuse reflection spectroscopy technology Applies to white granulated sugar to provide a reliable and robust quantitative analysis of the models,as well as for the design of high Signal to Noise Ratio spectrophotometry system of small-scale sugar-specific instruments of Near infrared diffuse reflection provide a basis.At first,it is based on the discrete wavelength points of MLR models of its optimization and stability analysis,including the stability and optimization analysis of the single-point regression models and the moving window interval MLR models.And then it is based on the continuous wavelength bands of PLS models of its optimization and stability analysis, including the stability and optimization analysis of whole region of direct PLS models and the Savitzky-Golay smoothing PLS models.Research on the stability of the Savitzky-Golay smoothing parameters and PLS factors.And using the moving window PLS method can choose the high Signal to Noise Ratio characteristics wavelength bands.The results show that:the method of dividing the calibration sample set and validation sample set has a good representation,objectivity and stability;the modle of using fixed interval combination discrete wavelength points has best prediction accuracy; Savitzky-Golay smoothing optimization mode can be greatly improved the predict results of models;Moving window PLS method can select the characteristics wavelength-bands of high Signal to Noise Ratio.Above accuracy of quantitative analysis can be used for the rapid detection of color value of white granulated sugar.