摘要
目的:建立一种新的模式识别方法:双指标等级序列模式识别法,该法兼有无监督聚类的客观性及有监督学习的明确分类特点。方法:根据正态分布,该法数学定义了一种质量等级标准,可在不同质量等级标准下进行模式识别,实现质量定量评价分级。根据优良等级相似样品序列,无须学习标准样品就可以同时对样品无监督聚类及分类,其中的相似样品自身作为分类标志,该法有效消除模式识别中的冗余信息,可以对一类中的样品进行更加精细的亚类模式识别能力。结果:对27种明目地黄丸及知柏地黄丸等浓缩丸样品无水乙醇提取物的红外指纹图谱进行了分析,明目地黄丸样品及知柏地黄丸样品分别形成各自的精细亚类,得到合理的模式识别结果。结论:双指标等级序列模式识别法具有准确质量定量评价分级能力。
Objective： A new pattern recognition method suitable for traditional Chinese patent medicine was established in this paper, which is named as the Dual index grade sequence pattern recognition. Method： In this method the quality gradation was defined mathematically relying on normal distribution. By this way samples can be clustered and classified depending on which quality gradation is wanted, and the grading samples quantitatively relative to quality can be performed simultaneously. Especially, the redundant information with respect to pattern recognition hiding in dual index sequences of samples can be removed effectively by applying the good grade sequences, which make the pattern recognition results accurate excellently. This approach possesses the advantages of both supervised classification and unsupervised cluster methods. Samples can be clustered and classified at the same time without any standard samples, and the operation is accomplished based on the good grade similar sequences themselves being as the classifying marks. Moreover, the subclasses in each class can be identified more subtly. Result： The infrared fingerprint spectra of extracts of 27 kinds of Mingmu Dihuangwan pills and Zhibo Dihuangwan pills samples extracted with ethanol were analyzed with the method proposed in this paper. The results showed that these pills can be classified in their subclasses clearly, respectively. Conclusion： The Dual index grade sequence pattern recognition is a new and effective one for identifying complex biological products made from complex herbal medicines.
出处
《中国中药杂志》
CAS
CSCD
北大核心
2008年第13期1543-1549,共7页
China Journal of Chinese Materia Medica
关键词
指纹图谱
序列
模式识别
聚类分析
明目地黄丸
知柏地黄丸
质量分级
fingerprint spectra
sequence
pattern recognition
cluster analysis
Mingmu Dihuangwan pills
Zhibo Dihuangwan pills
quality gradation
