近红外光谱预处理方法对毛/涤织物定量分析的影响

Analyzing of near infrared spectral preprocessing methods of the fiber content of wool/polyester blended fabric

为对比分析毛/涤混纺面料纤维含量的近红外光谱预处理方法,通过实验,对毛/涤混纺样品近红外光谱的采集、光谱预处理,特征值提取进行研究分析。光谱数据预处理方法包括标准化、导数、平滑和信号校正等。实验得出:选用均值中心化处理来提取谱图有用信息,Savitzky-Golay卷积平滑法消除噪声的影响;S-G一阶导来降低基线的漂移,在多种信号校正方法里选取多元散射校正方法,用该方法建模得出模型的预测平均绝对偏差(毛含量2.046 6,涤含量2.224 5)。近红外光谱预处理方法分析从光谱承载的众多信息中提取出了最有用信息,消除了样品表面的散射对近红外光谱的影响,降低了系统随机误差,进而提高了后续所建模型的稳健性和预测能力。

To analyze a variety of near infrared spectral preprocessing methods of the fiber content of wool/polyester blended fabric , through the experiment , collect the near infrared spectrum of samples, preprocess the spectrum and obtain the characteristic parameters. The preprocess methods include standardization, reciprocal, smoothness, signal correction. The experimental results show that compared with other pretreatment method effect, come to choose average centralized processing to extract the spectra useful information; Savitzky-Golay convolution smoothing method to eliminate the influence of noise; the S-G a derivative to lower baseline drift, in a variety of signal correction method to select multiple scattering correction method, the prediction mean absolute deviation of model （wool content 2. 046 6, polyester content 2. 046 6）. The analyzing of near infrared spectral preprocessing methods got the most valuable one from the mass information contained in the spectrum, eliminating the affection of of samples surface scattering on spectrum as well as system and random error, improving the stability and predictive ability of the model.