| Chinese herbal medicines(CHM) is our unique cultural heritage for thousands of years and it also is the basis of traditional Chinese medicine theory. With the rapid development of era, people prefer to use natural medicines. Chinese herbal medicines is more important as it’s natural, environmentally friendly, small for side effects, and so on. However, due to a wide variety of herbs and more planting areas, medicinal herbs that grows under different climatic will get the different compositions. Along with the rise of herbs’s price at market, people use some of the lower-priced herbs to replace the high price of herbs and this phenomenon has affected the normal development of the medicine market. Therefore, it’s now a hot research for the quality and species identification of Chinese herbal medicines. Many researchers have used a lot of the latest technology on medicinal discriminant analysis. But because of the many factors can affect the quality of herbs, these methods have achieved certain results. Due to the complexity and profession of the operation and the unity of the detection method, these methods are difficult to carry out promotion. In this paper, combined traditional methods with modern technology for herb identification, it identifies the quality of Chinese herbal medicines based on fusion information for odour and taste.The odour information and taste information are very important for traditional medicine identification. In this paper, firstly it obtains complete odour and taste information for Chinese herbal medicines through bionic olfaction system and bionic taste system. Secondly, it is identified in different fusion levels. This method not only overcomes incomplete information at detection method of the single type for herb, but also imitates the traditional methods. This is a scientific and standardized guide for traditional methods. The results are as follows:(1) In this paper, the five herbs are selected as the research object. They are Alpinia oxyphylla, Amomum cardamomum, atractylodes rhizome, atractylodes, angelica pubescens. In the best environment, it can be obtained for odour information and taste information throngh PEN3 E-nose and ASTREE E-tongue.(2) The research in this paper is to take full advantage of the learning ability for support vector machine (SVM) algorithm, and optimal parameter is found by a grid search method and K-fold cross-validation method and used for "one-to-one" multi-classification SVM. Odour information, taste information and fusion information are identified by the classifier. The result shows the selection of the kernel function will affect the recognition results, and recognition results which are based on fusion information perhaps are worse than others.(3) On the basis of the feature selection, a decision fusion based on D-S (Dempster-Shafer) evidence theory is proposed.lt frist analyzes the smell and taste of the Chinese herbal medicine through Support Vector Machine(SVM),and then gets the two kinds of features of that smell and taste as independent evidences by using the superiority of the evidence theory on combining the vague and imcomplete informations.The hard output of the SVM which changes into soft output through Platt Model is taken as Basic Probability Assignment(BPA). After implementing the evidence fusion on the combination rules, the recognition results is obtained according to decision rules finally, thus improving the stability and accuracy of the recongition.The result shows that the recognition rate of the decision fusion based on D-S evidence theory can be above 95%, which is better than the feature fusion based on SVM. |
PDF Full Text Request