中文摘要：

采用近红外漫反射光谱技术和偏最小二乘法（PLS）建立白砂糖色值的定量分析模型。用多元散射校正方法对光谱进行预处理，再用Savitzky-Golay平滑化方法对原谱、一阶导数谱和二阶导数谱进行处理。选取5个波段，每个波段分别采用原光谱、一阶导数谱、二阶导数谱。同时调整Savitzky-Golay平滑点数和PLS因子数，通过多次PLS数值实验比较，按照预测效果确定每个模型的最优平滑点数、因子数，再从中选优。结果表明，采用780～1100nm一阶导数谱的定标效果最好，模型的预测均方根偏差、相对预测均方根偏差分别为11-2，8．91％。780～1100nm可以代替近红外全谱波段（780-2500nm）得到好的定量分析效果，为设计小型专用近红外分析仪器提供依据。

英文摘要：

The calibration models for color value of white granulated sugar are constructed by near infrared diffuse reflection spectroscopy technology and partial least squares （PLS） regression. The spectra is pretreated by multiplicative scatter correction method, and makes use of Savitzky-Golay smoothing method to do smooth processing to the original spectra, the first derivative spectra and the second derivative spectra. 5 wave bands are selected, and adopting the original spectra, the first derivative spectra and the second derivative spectra respectively in each band. By adjusting Savitzky-Golay smoothing point number and the PLS factor number simultaneously, compared with many PLS computational experiments, it gets optimal smoothing point number and factor number for each model according to the prediction effect, and then selects the best one from them. The result shows that the prediction effect by using the first derivative spectra in 7804-1100 nm is best, and the root mean square error of prediction and the relative root mean square error of prediction for the corresponding model are 11.2, 8.91% respectively. It shows that 780-1100 nm band can replace the whole band of near infrared （780~2500 nm） and get good quantification effect, which provide base for designing minitype special near infrared analyzer.