| As Tibetan medicine plays a more and more important role in Chinese traditional medicine area, it has been an urgent thing to identify and evaluate Tibetan medicine’s quality. However, there are a series of problems like adulteration and imitation has arisen at present time in the Tibetan medicine market. Infrared spectroscopy has some merits like rapid, specialty and low cost. Because of these merits, it is replacing the common testing methods quickly. Based on some representative medlar samples, this paper confirmed the sample preparation methods and determination conditions through experiment, and obtained stable infrared spectrogram data. After that, the author tried to use computer to parse these data to establish a method to identify medlar. The main contents and achievements are summarized as follow:1. Collect some medlar samples from 18 different habitats and do necessary pretreatment(sun dried, kiln dried, grind and screening).2. Through experiment to determine medlar sample’s infrared spectrogram determination condition, then get medlar sample’s infrared spectrogram system atically and made an infrared spectral interpretation for those eighteen representative samples. Through the contrast and analysis of these samples’ infrared spectrogram, this paper obtains the conclusion that the spectrograms have some differences in the fingerprint region(1500cm-1-500cm-1). Besides, different-habitat medlars’ infrared spectrograms and quality are almost the same. The main differences are mainly refers to the concentration and varieties of various chemical components in fingerprint region which mainly depends on the medlar samples’ growth and geographical environment.3. Use a variety of pretreatment methods to pretreat all these 18 kinds of medlar samples’ infrared spectrograms then compare these different methods’ pretreatment results. Through this process, find that the wavelet transform is the most ideal pretreatment method.4. Establish the basic principle and assessment index of wavelet transform--based pretreating, choose the appropriate mother wavelet function and threshold, confirmed the denoising and threshold compression value. Meanwhile, compress the habitat infrared spectrogram’s data length from 1868 to 241. Through this way, this study realized to let the root-mean-square error(RMS) achieves the 10-4 order of magnitudes. This indicates that the compressed infrared spectrograms can be analyzed by computer.5. The habitat identification adopts principal component analysis and cluster analysis. Between them, the variance ratio of the first three principal components are about 99.407%(λ1 = 184.64,λ2 = 47.267, λ3 =7.6641), which shows that the first three principal components can reserve 99.407 percentages of infrared spectrum data matrix. Through this way, these medlar samples can be divided into three types just as cluster analysis does. Principal component analysis and cluster analysis results are consistent. On the one hand, it verified that principal component analysis results are reliability; on the other hand, luster analysis can be viewed as a supplement for medlar’s origin identification. In all, both principal component analysis and cluster analysis can be used to identify medlar’s origin.6. The Quality identification adopts similarity analysis and grey relational analysis, which shows that the included angle cosine data(cosαst) concentrate in the area between 0.99 and 0.95 However, the correlation coefficient(Rst) and grey correlation coefficient data scattered over the area between 0.9 and 0.6. So, the quality identifica- tion can based on the similarity data and we can get a satisfied result.7. Use partial least square discriminant analysis(PLS-DA) to establish these 18 different habitats medlar samples’ pattern recognition system, and choose 15 samples as the modeling training set to distinguish the other three samples. The recognition rate is 100%. |
PDF Full Text Request