利用便携式近红外光谱仪对376个涤/棉混纺织物进行研究, 利用定量分析模型中的偏最小二乘法(partical least squares, PLS)作为校正方法, 结合涤/棉混纺织物中涤、 棉含量设定的定性鉴别系数, 建立了涤/棉混纺织物的半定量-定性分析校正模型。 该模型对涤/棉混纺织物进行定性鉴别的同时得出其相对含量, 分析结果具有半定量性质。 在建模过程中, 采用Savitzky-Golay导数法, 消除噪声和基线漂移对光谱的影响, 并研究了波段选择和不同预处理方法对定性校正模型的影响。 纯棉的主要吸收峰位于1 400~1 600 nm, 纯涤的主要吸收峰位于1 600~1 800 nm, 随着涤或棉含量的增加, 其相应的吸收峰强度增强, 因此, 建模波段以涤、 棉主要吸收峰区间为基本波段, 进行双向扩展, 得到最佳波长区间1 100~2 500 nm(相关系数0.6, 波点数934)。 利用所建校正模型对验证集样本进行预测, 结果表明, 在1 100~2 500 nm处, 预处理方法为Savitzky-Golay导数、 多元散射校正与均值中心化相结合时, 该模型评价参数较佳, 其中RC(校正集相关系数)0.978, RP(验证集相关系数)0.940, SEC(校正标准差)1.264, SEP(预测标准差)1.590, 样品预测正确率达93.4%。 表明该定性分析校正模型能够较好地对涤/棉混纺织物进行半定量-定性预测。

A series of 376 cotton-polyester (PET) blend fabrics were studied by a portable near-infrared (NIR) spectrometer. A NIR semi-quantitative-qualitative calibration model was established by Partial Least Squares (PLS) method combined with qualitative identification coefficient. In this process, PLS method in a quantitative analysis was used as a correction method, and the qualitative identification coefficient was set by the content of cotton and polyester in blend fabrics. Cotton-polyester blend fabrics were identified qualitatively by the model and their relative contents were obtained quantitatively, the model can be used for semi-quantitative identification analysis. In the course of establishing the model, the noise and baseline drift of the spectra were eliminated by Savitzky-Golay(S-G) derivative. The influence of waveband selection and different pre-processing method was also studied in the qualitative calibration model. The major absorption bands of 100% cotton samples were in the 1 400～1 600 nm region, and the one for 100% polyester were around 1 600～1 800 nm, the absorption intensity was enhancing with the content increasing of cotton or polyester. Therefore, the cotton-polyesters major absorption region was selected as the base waveband, the optimal waveband (1 100～2 500 nm) was found by expanding the waveband in two directions (the correlation coefficient was 0.6, and wave-point number was 934). The validation samples were predicted by the calibration model, the results showed that the model evaluation parameters was optimum in the 1 100～2 500 nm region, and the combination of S-G derivative, multiplicative scatter correction (MSC) and mean centering was used as the pre-processing method. RC (relational coefficient of calibration) value was 0.978, RP (relational coefficient of prediction) value was 0.940, SEC (standard error of calibration) value was 1.264, SEP (standard error of prediction) value was 1.590, and the samples recognition accuracy was up to 93.4%. It showed that the cotton-polyester blend fabrics could be predicted by the semi-quantitative-qualitative calibration model.