다양한 다변량 거리측정법에 따른 군집분석에 의한 황금 원산지 구분

Abstract

NIR(Near-Infrared) spectroscopy is a nondestructive technique and gives fingerprint-type information. Its scanning speed is much fasten compared to other analytical techniques, and usually, minimum sample preparation is required. Particularly, it has been well studied in agricultural and food industry because of fast analysis. Recently, it became important to discriminate the geological origins of natural products for pharmaceutical uses. In this paper, a method of pattern recognition basically based on a form of cluster analysis is described for the discrimination of geological origin of Scutellaria Radix. When 2nd derivative of NIR spectra were used for the calculations of multivariate distances, the results of geological discrimination of the crude drug was successful without any misclassification. The result of successful and perfect discrimination of geological origins of the samples was achieved using several multivariate distance measures, such as Euclidean distance(2-norm), Minkowski distance with norms of 0.5, 1.5 or 3, cosine distance, city block distance(l-norm), or Chebychev distance(oo-norm). With correlation distance, there were a few cases of wrong classification results. On the contrary, with reflectance spectra or 1st derivative spectra, the discrimination was not successful and some distance measures such as Mahalanobis distance or Standardized Euclidean distance were not appropriate for this kind of cluster analysis.