Research On Classification Of Green Tea Based On Computer Vision And Electronic Tongue

The conclusions of tea evaluation often were likely to be different. For an operator, the factors such as physical,experience and circumstances can have effects on the results; And for different operators, sometimes they can not draw the same conclusion with the same sample, finally the accuracy of the evaluation of tea quality was swayed. In this dissertation, the evaluation for tea quality including the appearance and inherence was studied by using image processing and electronic tongue.The main contents of the study were as follows:1．There were twelve features were quantitatively extracted from the images ofgreen tea. The color features were composed of the Mean of Hue((?)),Intensity((?)),Saturation((?)) and the Variance of them(σ_H,σ_I,σ_S); The texture features comprisedstatistical moment and Gray Level Co-occurrence Matrix, including Mean( m),Standard Deviation(σ),Smoothness( R),Third Moment(μ₃),Uniformity( U),Entropy( e) and Energy( Q₁),Homogeneity( Q₂),Contrast( Q₃),Correlation( Q₄) . Byusing Pattern Recognition, all the features had been processed with Bayesian Theory to classify and come to a conclusion that combining with color and Gray Level Co-occurrence Matrix features can discriminate different class green tea. With the 120 samples, it's result showed that the identification rate of training set and cross-validation set all reached 100%.2．Based on electronic tongue, with four grades of roasted green tea liquor, the features of sensors' response values were collected to be discriminated. After principal component analysis (PCA) pre-processed, the independent components were extracted from the original data. With three supervised pattern recognitions, Linear Discriminant Analysis (LDA), K-Nearest Neighbors (KNN) and Back Propagation Artificial Neural Network (BP-ANN) were applied to build discriminating models. The performance of three pattern recognition methods on electronic tongue data was compared. The result showed that among the three built models, BP-ANN model was the best for discriminating the four roasted green teas, the optimal principal components factors (PCs) equals to 5, and both discriminating rates equal to 100% in training and validation set. For more quickly and efficiently discriminating the green tea, this research can provide some meaningful ways to assist the operator of tea testing to make an objective description and evaluation, moreover, it can also promote the standardization of the market of tea and make the marketplace come to maturity.