SIMCA采用PCA模型参数和F检验构造计算T2i/T2ucl和Si/Q统计量作为样本分类的新属性, 并计算待测样本到各类主成分空间的欧式距离作为判别类别的依据, 是一种最常用和优秀的光谱分类方法。 但是, 在Q对T2作图平面上, 以欧式距离确定的样本分布范围是一个圆, 多数情况下并不一定能符合实际样本分布规律。 本文在分析了SIMCA理论缺陷的基础上, 提出了一种新方法, 即用马氏距离代替欧氏距离作为判别依据来判断样本的类别。 并设计了采用红外光谱判别组分比例很接近的掺假食用油样本的实验, 以及用近红外光谱判别相近皮毛样本的实验。 用调和比5%～8%的食用油红外光谱PCA模型, 分别以马氏距离和欧式距离计算出其样本的分布范围, 结果表明马氏距离的分类与识别能力更强。 新方法和SIMCA对动物皮毛样本的正确识别率分别为87.5%和75%, 对比例相近的食用油调和油的正确识别率分别为65%和55%。 结果表明新方法对化学组成差异微小的样品分类精度明显优于SIMCA。

In the SIMCA, the parameters of PCA model and F test are used to construct T2 and Q for classification, and Euclidean distance is used to determine the range of sample distribution of the model. Since the range which is defined by Euclidean distance is a circle in the plane of T2 vs Q, the boundary of actual samples which distributes in some directions and irregular space cannot be presented accurately. Besides, SIMCA is still inaccurate for classification and identification in theory. Therefore, a new multivariate classification and identification method was proposed using Mahalanobis Distance instead of Euclidean distance in this paper. Experiments of infrared spectra of blending edible oils and near infrared spectra of animal furs were designed to compare the performance of the new method and SIMCA. The recognition rates of the new method and SIMCA for three kinds of furs are 85.5% and 75%, respectively. The recognition rates of the new method and SIMCA for two classes of blending edible oils are 65% and 55%, respectively. It has shown that the new method is superior to SIMCA in the performance of discriminating the different materials with a small difference in their chemical composition.