Study On The Digital Tea Blending Tea Of Keemun Gongfu Black Tea Based On NIR And HSI Technology

Blending is one of the important and key steps in export tea processing.However,in the present situation blending technique completely depends on artificial qualitative analysis(Sensory evaluation).To realize the intelligentize of the tea blends,the tea blends samples were preparation to acquire near infrared spectroscopy and hyperspectral imaging.The contents of amino acids and caffeine from samples were detected by high-performance liquid chromatography,and the tea polyphenol was detected by GB/T 8314-2008.The spectral and textural features were extracted from hyperspectral data to discriminate the relative proportions of the raw material constituents of each blend.Data fusions combined with the contents of amino acids,caffeine and tea polyphenol were used to predict the relative proportions of the raw material constituents of each blend.The results showed as the following:(1)Near infrared spectroscopy(NIR)combined with contents of chemical index build the models for estimating the relative proportions of the raw material constituents of each blend.Comparing different models with the spectral features,the optimal quantified model is filtered out.Experimental results showed that the amino acids,caffeine and polyphones optimal modle offers its Rc in 0.9486,0.9899 and 0.9863 respectively by Si-PLS method.The results showed that the Near infrared spectroscopy combined with contents of chemical index can be used to predict the relative proportions of the raw material constituents of each blend.(2)Near infrared spectroscopy(NIR)combined with back propagation-artificial neural networks(BP-ANN)build the models for estimating the relative proportions of the raw material constituents of each blend.Cross-validation method was used to select the principal component number by cross-validating the training set samples one by one.When the principal components number was 11,the result is best.The discrimination rate is 77.6%.Single spectral information is not able to solve more complex tea blending problems.(3)Modeling to discrimination of the relative proportions of the raw material constituents of each blend based on spectra data and texture features.Discriminant models were built based on the textural features from first two principal component images and five dominant wavelengths of images,respectively.The results showed that The LS-SVM model based on data fusion gave the best results with high correct discrimination rate of 94.5%.The results implied that data fusion combined with LS-SVM has the capability of discriminating the relative proportions of the raw material constituents of each blend.It provided research basis for the digitization and production standardization of tea blends.