| Currently, the sensory evaluation of tea quality is conducted by a panel of trained tasters and chemical analysis methods. However, such kind of sensory evaluation method leads to experimental results with randomness, destructive, poor repeatability. Chemical analysis methods are extremely time-consuming, expensive and destructive. Hence, some fast and non-destructive detection technologies such as computer vision technology, hyper-spectral imaging, color meter, electronic tongue and near-infrared (NIR) spectroscopy, were used in this research to evaluate the sensory quality of tea. The famous green tea "Biluochun" (Camellia sinensis (L.)) was regard as the object for this research. The calibration models were built based on the relationship between Tea Tasters'scores and the results from the fast and non-destructive dectection technologies. Therefore, the green tea quality can be predicted quantitatively by the developed calibration models. The results of this work can provide a reliable method of assessing the tea quality in order to standardize the marketing of tea. The main points are as follows:(1) The quantification methods for the sensory quality based on appearance of the famous green tea by computer vision technology and hyper-spectral imaging technology were studied. A computer vision system was used to capture the visible images of dry teas samples.12 color features and 28 texture features were extracted from each image. Partial least squares (PLS) and back propagation artificial neural network (BP-ANN) were performed to develop the calibration models based on the relationship between Tea Tasters' scores for the appearance of the tea samples and the extracted feature variables from the images. Experimental results showed that the performance of BP-ANN model was better than that of PLS. Root mean square error of prediction (RMSEP) and correlation coefficient (Rp) of BP-ANN for the prediction set were 2.396 and 0.937, respectively. A hyper-spectral imaging system was used to obtain the hyper-spectral images of dry teas samples. Principal component analysis (PCA) method was applied to select the optimal characteristic wavelengths, and three gray images with the optimal characteristic wavelengths were gotten from each hyper-spectral image. Then, two color features and 28 texture features were extracted from each selected gray image; and therefore,90 features were obtained for each sample. PLS and BP-ANN methods were also employed to develop the calibration models based on the relationship between Tea Tasters'scores for the appearance of tea samples and the extracted feature variables. Experimental results showed that the performance of BP-ANN model was better than that of PLS, where RMSEP was 3.611 and Rp was 0.859 from the BP-ANN for the prediction set. The performances of the developed models by the two technologies were compared. The results showed that the models based on the computer vision technology had a better performance.(2) The quantification methods for the sensory evaluation of infusion color of the famous green tea based on color measurement technology were studied. A color meter was used to measure the colorimetric values of tea infusion samples. Stepwise regression and PCA methods were respectively applied to extract colorimetric features. PLS was used to build the calibration models based on the relationship between Tea Tasters'scores for the infusion of tea samples and the feature variables extracted by stepwise regression and PCA, respectively. Experimental results showed that the model based on the feature variables by PCA got better result. Then, PLS and BP-ANN methods were used to develop the models based on the features by PCA, respectively. The results showed that the performance of BP-ANN with RMSEP=2.505 and Rp=0.816.was better than that of PLS for the prediction set(3) The quantification methods for the sensory taste quality of the famous green tea based on chemical analysis instruments, near infrared spectroscopy and electronic tongue were studied. HPLC and spectrophotometry methods were used for the determination of main 10 taste components such as tea-polyphenols, amino acid, caffeine, GA, EGC, C, EGCG, GCG, ECG and the total catechins content. PLS and BP-ANN methods were applied to develop the calibration models based on the relationship between Tea Tasters' scores for the taste of tea samples and the 10 taste components, respectively. Experimental results showed that the performance of BP-ANN model was better than that of PLS, where RMSEP and Rp were 2.553 and 0.869, respectively. NIR spectra were acquired in the reflectance mode using the Antaris II Near-infrared spectrophotometer with an integrating sphere for the tea samples. PCA was employed to extract the spectra feature variables. PLS and BP-ANN were also applied to develop the calibration models based on the relationship between Tea Tasters'scores for the taste of tea samples and the feature variables extracted by PCA. The results revealed that the performance of BP-ANN model with RMSEP=2.104 and Rp=0.916 was better than that of PLS for the prediction set. At the same time, electronic tongue was applied to get the sensor response values of tea infusion. PLS and BP-ANN were also applied to develop the calibration models based on the relationship between Tea Tasters'scores for the taste of tea samples and the sensor response values of the electronic tongue. It was revealed that the performance of BP-ANN (RMSEP= 1.913, Rp=0.932) was better than that of PLS. The performances of the developed models using the above three technologies were compared. The results showed that the models based on the electronic tongue technology had the best performances, while the performance of the model based on chemistry instrument methods was not satisfied.(4) The quantification methods of the sensory quality of infused tea leaves for the green tea based on computer vision technology and hyper-spectral imaging technology were studied. The computer vision system was used to capture the visible images of the infused leaves.12 color features and 28 texture features were extracted from each image. PLS and BP-ANN were employed to develop the calibration models based on the relationship between Tea Tasters'scores for the infused leaves of tea samples and the extracted feature variables. Experimental results showed that the performance of BP-ANN was better than that of PLS, where RMSEP and Rp of BP-ANN in the prediction set were 2.496 and 0.863, respectively. Also, the hyper-spectral imaging system was used to get the hyper-spectral images of the infused leaves. PCA was applied to select the optimal characteristic wavelengths, and three gray images with characteristic wavelengths were taken from each hyper-spectral image. Then, two color features and 28 texture features were extracted from each selected gray image;and 90 features were obtained for each sample. PLS and BP-ANN were used to develop the calibration models based on the relationship between Tea Tasters'scores for the infused leaves of tea samples and the extracted feature variables. Experimental results showed that the performance of BP-ANN with RMSEP=2.626 and Rp=0.846 was better than that of PLS in the prediction set. The performances of the developed models using the above two technologies were compared. The results showed that the models based on the computer vision technology were better.(5) The quantification methods of the taste quality chemical evaluation of the famous green tea by NIR spectroscopy and electronic tongue were studied.Chemical evaluation was the reference measurement used to measure total taste scores of green tea's infusion. Synergy interval PLS (siPLS) and genetic algorithm (GA) were used to select the optimal feature variables from the NIR spectra. The models based on the relationship between the total taste scores and the selected feature variables from the NIR spectra were developed. The optimal model was achieved with Rp=0.8908, RMSEP=4.66 for the prediction set when 38 spectra variables and six PLS factors were included. In addition, PLS and least squares support vector machines (LS-SVM) methods were applied to develop the calibration models based on the relationship between the total taste scores and electronic tongue measurements. The results showed that the LS-SVM is superior to the PLS model. RMSEP and Rp values of LS-SVM for the prediction set were 0.906 and 4.077, respectively.The main aim of improving the objectivity and definiteness of the sensory evaluation of green tea quality has been achieved. The results in this research can provide references for establishing a scientific, rational and consistent evaluation standard in tea industry, and also can provide foundation for developing instruments for quantification of tea quality.
|
PDF Full Text Request