Research On Pattern Analysis Methods Based On Multiple Graph Presentation Of Traditional Chinese Medicine Fingerprint

The study on the modernization of Traditional Chinese Medicine (TCM)is a hot topic.The fingerprint technology is a powerful tool for modernizing the TCM, and it can be usedto identify the authenticity, control the quality and evaluate the safety and effectiveness.We process the TCM fingerprint by means of computer usually, analyzing their chemicalcomposition measured values and the related pharmacological effects, and finding out theinherent laws quickly and accurately as an indicator of quality control of TCM. Theintegrity of the fingerprint data is essential to the research because the TCM is synergisticeffect by multiple components. However, some researches break the integrity in analysisprocess, and some researches ensure the data integrity but with poor intuitive and hard tounderstand in the current study. Therefore, how to realize the visualization of the analysisprocess on the basis of ensuring the integrity of TCM data is an important issue in theTCM fingerprint research.Aiming at the data process problems of the TCM, in this thesis, a novel visualizationmethod of pattern analysis based on the multiple graph representation theory wasadvanced, which will help to ensure the unity of integrity and obscurity of the TCM andprovided a novel information process means for the identification and evaluation of theTCM. Three basic problems were focused on in the work of this thesis, multiple graphrepresentation and feature extraction of multivariate data, the construction of classificationmodel in radar plot of the TCM fingerprint data and the construction of the standardfingerprint.Firstly, the multivariate graph representation method of the multivariate data and thefeature extraction method based on the multivariate graph principle were studied. Thesorted overlap coefficient matrix was proposed to eliminate the variables with biggervariance and little classification information. The traditional scatter matrix was optimizedto separate the viables with similar sorted mean and whole mean. The sample wasrepresented by the distance between the sample and the class hyperplane and the distancewas as feature for classification. The application of Human-Computer Interaction (HCI)in feature extraction was studied. Experiments demonstrated the feasibility and advantages ofthese methods.Secondly, the classification of the TCM fingerprint based on the principle of radardiagram representation was studied. The center feature was expanded and the adjacentamplitude ratio was optimized based on the representation and feature extraction of theradar plot of multivariate data. Aiming at the classification problem of the TCMfingerprint, the center feature peak and the multi-layer radar plot representationclassification model were proposed firstly. Experiments had achieved good results.Thirdly, the construction of the standard fingerprint was studied. Aiming at theproblems of traditional methods for building the standard fingerprint, based on theclustering theory, a novel standard fingerprint that uses the clustering rules of themulti-parameter distance was put forward. Because classification is the purpose of themethod, the result is no longer a characteristic fingerprint or library with traditionalmeanings, but a classification standard. Experiments analyzed the feasibility andsuperiority.Finally, using the near infrared spectroscopy, the fingerprints of the Radix Puerariae(pure and adulterated goods)and ginseng (different types)were obtained. The distinction ofthe authenticity of radix Puerariae and the species of ginseng were studied. Theexperimental results verified the feasibility and effectiveness of the proposed visualizationanalysis method of TCM fingerprint.