可见-近红外光谱用于鉴别山羊绒与细支绵羊毛的研究

Identification of Fine Wool and Cashmere by Using Vis/NIR Spectroscopy Technology

近红外光谱作为快速、无损的检测技术,近年来在国内外越来越受到广泛关注。针对山羊绒与细支绵羊毛的可见/近红外光谱的特点,提出了应用主成分分析(PCA)结合人工神经网络(ANN)进行山羊绒与细支绵羊毛的鉴别,并建立了羊毛、羊绒分析模型。应用可见/近红外反射光谱获取山羊绒与细支绵羊毛的光谱曲线,利用主成分分析对原始光谱数据进行处理,根据主成分的累计贡献率99.8%选取主成分数6,并将所选取的6个主成分作为三层BP神经网络的输入。通过定标集样本对BP神经网络进行训练,用优化的BP神经网络模型对预测集样本进行预测。实验结果表明,16个未知样本的鉴别全部正确,表明可见/近红外光谱结合主成分分析和神经网络技术对山羊绒与细支绵羊毛进行快速鉴别是可行的。

As a rapid and non-destructive methodology, near infrared spectroscopy technique has been attracting much attention recently. The present study applied Vis/NIR spectra to the identification of cashmere and fine wool fiber. Cashmere and fine wool are resemble in superficies, but they differs in diameter, height, thickness, angle of inclination, and marginal morphology of surface scale. Although researchers both at home and abroad did a lot researches and experiments to distinguish fine wool from cashmere, the resolution of cashmere and fine wool is still not satisfactory, and it is always a challenging task to differentiate and recognize fine wool and cashmere. This paper presents an automatic recognition scheme for the fine wool fiber and cashmere fiber by Vis/NIR spectroscopy technique, aiming at the characteristics of Vis/NIR spectra of cashmere and fine wool. One mixed algorithm was presented to discriminate cashmere and fine wool with principal component analysis （PCA） and artificial neural network （ANN）. Preliminary qualitative analysis model has been built. Vis/NIRS spectroscopy diffuse techniques were used to collect the spectral data of cashmere and fine wool, and two kinds of data pretreatment methods were applied： the standard normal variate （SNV） was used for scatter correction. Savitzky-Golay with the segment size 3 was used as the smoothing way to decrease the noise processed. Following the pretreatment, spectral data were processed using principal component analysis, 6 principal components （PCs） were selected based on the reliabilities of PCs of 99. 8%, and the scores of these 6 PCs would be taken as the input of the three-layer back-propagation （BP） artificial neural network （BP-ANN）. The BP-ANN was trained with samples in calibration collection and predicted the samples in prediction Collection were predicted. Experiments demonstrate that the system works quickly and effectively, and has remarkable advantages in comparison with the previous systems. The result indicated that a model had been built to discriminate cashmere from fine wool using Vis/NIR spectra method combined with PCA-BP technology.