Clustering analysis for parted band based on infrared spectra of tea

The standard mid-infrared spectra of fifteen kinds of tea that respectively belongs to slight fermentation (green tea, yellow tea, white tea), moderate fermentation (oolong tea) and deep fermentation tea (black tea, dark tea) are measured by Fourier transform infrared spectroscopy. By the visual analysis on the spectra it is found that they are strongly similar, so it is difficult to distinguish the different spectra directly. The main characteristics of spectra are obtained by informatics method of feature extraction, and then the scientific method of clustering analysis for teas belong to different fermentation degree is found. In order to further improve the validity of clustering result and emphasize the contrast of character parts, the zonal analysis of multiband for the spectra is done. The results show that using feature extraction for spectra in the wavenumber range of 1800 to 800cm^-1, the clustering effect is better than that in wavenumber ranges of 3700 to 2400cm^-1, 3700 to 2400cm^-1 plus 1800 to 800cm^-1, 1300 to 600cm^-1 and all wavenumber range of 4000 to 400cm^-1. The above results indicate that the wavenumber range of 1800 to 800cm^-1 is the primary spectral region of reflecting characteristic of tea, which doesn´t drop the primary information of reflecting fermentation feature of tea, and almost doesn´t include the interferential information of nonfermentation feature of tea.