选用不同厂家的红薯淀粉、马铃薯淀粉和玉米淀粉共112个样品,利用短波近红外光谱技术对淀粉种类进行鉴别。分别采用马氏距离判别法、C-支持向量机(C-SVM)、ν-支持向量机(ν-SVM)建立淀粉种类鉴别的短波近红外光谱模型;并对比多元散射矫正、平滑、一阶微分、二阶微分等多种预处理方法后的建模结果。结果表明:同时使用平滑、多元散射矫正、一阶微分3种预处理方法后,ν-SVM分类模型的效果最佳;训练集交叉验证正确率为100%,测试集正确率也达到100%。该模型快速准确无损的鉴别淀粉种类是可行的。

A method of starch category analysis was developed using Short-wave Near-infrared(NIR) spectroscopy.All 112 samples were obtained from different manufacturers of sweet potato starch,potato starch and corn starch.The Short-wave NIR models were established using mahalanobis distance discriminant,C-Support vector machine(C-SVM) and ν-Support vector machine(ν-SVM).The various different pretreated methods(multiplicative scatter correction(MSC),smooth,first-derivation and second-derivation of spectra data were applied.The results indicated that ν-SVM got the best results after MSC,smooth and first-derivation pretreatments.The correct ratio of the training set and the testing set is 100% and 100%.The results showed the method for simultaneous,non-destructive analysis in determination of starch category is reliable.