Identification Of Chinese Medicine Based On Bionic Olfactory And Kernel Method

Traditional Chinese medicines (TCMs) are the most important material carrier of Chinese medicine, which has been responsible for the treatment and health fitness of Chinese people for thousands of years. Even in today's western medicine society, Chinese medicine is still many Chinese people's and overseas Chinese people's important medical and health way. Authentic TCMs are those with Chinese characteristics, recognized traditions, and from specific areas. However, due to a wide range of medicines and the complexity of sources, shoddy goods of many authentic TCMs exist on the market, leading to confusion in the medicine market, so that people's health and property suffers damage. Therefore, the research on the controllability of the quality of authentic TCMs has been a hot topic in Chinese medicine sector. "Man-smell" is one of the traditional methods to identify TCMs.This paper presents bionic olfactory technology to simulate human nose, and creates a new method to measure and identify authentic TCMs.Bergamot, kaempferol, Motherwort, ginger, Bupleurum, mint, Amomum are selected as research objects, and bionic technology and nuclear methods are utilized to achieve classification and identification of authentic TCMs. The results are as follows:(1) Chinese herbal medicine samples are measureed by a portable electronic nose (PEN3). Radar plots of TCM samples are drawn so as to reflect the volatile characteristic of each variety intuitively. Then the relationships between the response curves or radar plots and TCM samples with different varieties, harvest date and growing areas are studied.(2) According to the trend and changes of each sensor's response signals, three static characteristic parameters and two dynamic characteristic parameters are generated to construct the original feature vector of each sample. Static characteristic parameters include maximum response, average value and standard deviation during the entire sampling process; dynamic characteristic parameters include average value during the steady sampling process and the response value of each sensor while having the maximum variance.(3)According to the theory of nuclear methods,quadratic kernel function,order polynomial kernel function, radial basis function kernel functions are chose, and three groups of related parameters are respectively set.And then the processing results of each kernel function based on the same data and different sets of parameters are analysed, which and the processing results by PCA+LDA method are comparised. In general, kernel-based method is better than PCA+LDA method. The results by Fisher discriminant analysis based on radial basis (Gaussian) kernel function (when the parameter is 50) are the best, followed by Fisher discriminant analysis based on the quadratic kernel function (when the parameter is 20),but the results by Fisher discriminant analysis based on d order polynomial kernel function are not ideal.(4) Fisher discriminant analysis based on radial basis (Gaussian) kernel function (when the parameter is 50) is used to classify and identify TMCs with with the same harvest date, different growing areas and different growing areas. The classification rate of each measured sample reachs 100%.It works out that it's possible to identify authentic TCMs accurately by a portable electronic nose.