The Study Of Algorithms And Applications On Evaluating Traditional Chinese Medicine Fingerprint

Traditional Chinese medicine (TCM) is a complex mixture of a multitude of chemical components. Due to the special feature of TCM, it is difficult to achieve a standardization and modernization of TCM`s quality control. The TCM fingerprint is considered to be a comprehensive method for quality control of TCMs. There are key problems about processing TCM fingerprint, which should be researched using by modern digital singal processing technic, math statistic methods and artifical intelligent algorithms.For TCM fingprint, most reaseaches focuses mainly on two directions, one is to create fingerprint, and the other is to process fingerprint signals. At present, there has been a wide research in creating fingerprint performed by pharmic experts. However, until now there haven`t algorithms of processing fingerprint signals, which are precision and dependable. Until now there haven`t a software system of processing fingerprint signals, which are cheap and easy-to-maintain throughout inland and overseas. Therefore, it is significant in theory and application to research algorithms and develop system in processing TCM fingerprint. This dissertation begins the research work from the shortcomings of former researchs or the methods that are never concerned. By adding process information, using simple and effective data preprocessing method, using new modeling method with good generalization ability and building a hybrid model with better performance, some new fingerprint processing methods are put forward to improve experimental precision of TCM fingerprint similarity. In order to get better experimental results, various kinds of theorys and methods should be integrated to dig out useful information from the original data.In the research of this dissertation, the advantages of the SUN YAT-SEN university and our team are combined, and supported by the Science and Technology Project of Guangzhou, now a great deal of works have been done in processing TCM fingerprint. The results we got have achieved certain efficiency in real applications, but farther research is needed to improve the calculating precision and the wide application.This dissertation concentrates on the research work listed below and achieved some creative results: 1. Improve and implement a new method to correct the baseline drift of TCM fingerprint. The method is to split the baseline of the TCM fingerprint signal via the second-generation wavelet transform. Although the first generation wavelet transform has been successfully applied to the signal processing, the complexity of the calculation is troublesome. To overcome this problem, we apply the second-generation wavelet transform technology to processing the fingerprint signal and effectively remove the baseline from the original signal.2. Improve and implement a new method to correct correct retention time shift of TCM fingerprint. The method is to correct retention time shift by the local least squares fitting. The retention time shift is the other important issue of preprocessing the TCM fingerprint signals. The ordinary least squares fitting algorithm is not robust enough to handle the problem of retention time shifting because of the complicated multi-component signals. The authors employ a practical linear fitting method, i.e. the local least squares fitting, to deal with this problem.3. Put forward and implement a new algorithm to establish the common pattern of TCM fingerprint. Establishing the common pattern of TCM fingerprint signals is an important problem for the evaluation in herbal medicine field. This alogorithm deals with the application of support vector machine regression combined with discrete wavelet transform in creating the common pattern of fingerprint signal. The original signal was first decomposed to different scales by discrete wavelet transform constructed by lifting scheme, and then the decomposed signal components were approximated by different kernel function of support vector machine regression.4. Put forward and implement a new algorithm to calculate the similarity of TCM fingerprint. Firstly the basis of similarity is accounted by angle of cosines, then the overlap rate of curve made by relative area or common peaks of fingerprint is get, finally combing the two coefficients to calculate similarity of fingerprint. The calculated similarity with the proposed method can express the global similarity of fingerprint and the special common peaks of fingerprint. And this method is simple to be coded.5. Put forward and implement a comprehensive evaluation method of TCM fingerprint. A new method of comprehensive evaluation of the similarity degree of TCM fingerprint signal, developed by the authors of this paper based on a combination of the fuzzy neural network and the genetic algorithm. Fuzzy membership functions are obtained by using Radial Basis Function neural network, and then genetic algorithm is applied to train fuzzy RBF neural network. The trained fuzzy neural network is used to evaluate the similarity of fingerprint signals.6. A new software system had already been realized based on the object oriented program (OOP) designing language and database. The real-life samples of TCM fingerprint proved by the Morden TCM Quality Reasearch Center of SUN YAT-SEN University are applied to the introduced methods above, and the experimental results are discussed, showing the validity of the proposed approaches.All of the aboved methods have made a certain effect to improve the accuracy of TCM fingerprint singals processing. Although these methods are put forward based on the research of digital singal processing and pattern recognization. They also have referecne to the experience of experts in TCM field. The contribution of this dissertation plays important role in enhancing, enriching and deepening the TCM fingerprint processing technology, and the software system will also have a wide application in the future.