利用可见/近红外反射光谱无损检测小米的粘度

目的建立适用于小米粘度无损检测的可见/近红外反射光谱法。方法使用光谱仪获取小米在367～1020 nm波段范围内的漫反射光谱,采用多元散射校正法(multiple scattering correction,MSC)和一阶导数法(first derivation,1~(st)-D)对原始反射光谱进行处理,并且使用主成分分析确定最佳主成分数,建立小米粘度判别模型,使用全交叉验证法进行模型验证。结果使用原始反射光谱、MSC处理光谱和1~(st)-D处理光谱,分别提取了6、12和12个主成分,建立的峰值粘度模型RCV在0.86以上,对验证集的预测结果 Rp在0.82～0.86之间;而使用1~(st)-D处理光谱提取12个最优主成分,建立的模型可较好地预测小米粘度的破损值,RCV为0.8573,对验证集的预测结果 Rp为0.8309。结论该方法适用于小米粘度的无损检测,为小米加工品质的快速检测提供一定的理论支持。

Nondestructive detection of viscosity of millet by visual/near infrared reflectance spectroscopy

Objective To establish a method for nondestructive determination of the viscosity of millet by the visual/near infrared reflectance (VIS/NIR) spectroscopy. Methods The VIS/NIR reflectance spectrum of millet in 367~1020 nm was collected by spectrography. The spectrum was pretreated using multiplicative scattering correction (MSC) and first derivative (1st-D). The best number of principal components were determined by principal component analysis. Then the calibration models of millet viscosity were established and the full cross validation were used to examine the effect of model. Results Six, 12 and 12 principal components were extracted by using the original reflectance spectra, MSC spectra and 1st-D spectra. The established peak viscosity model Rcv were more than 0.86, and the predicted result of the verification set RP were between 0.82 and 0.86. The 1st-D spectrum was used to extract the 12 optimal principal components. The established model could predict the damage value of millet viscosity better, Rcv was 0.8573, and the prediction result of the verification set RP was 0.8309. Conclusion This method is suitable for nondestructive testing of millet viscosity, and provides some theoretical support for the rapid detection of millet processing quality.