A Method Of Distinguishing Tea Quality Based On Hyperspectral Imaging

Tea is one of the three most popular drinks in the world.The variety and quality of tea have great influence on its taste.The traditional method of distinguishing tea quality mainly depends on the combination of human panel test and wet chemical analysis,which is time-consuming and laborious.Therefore,it is of great significance to study the method of rapid tea quality identification.The research content of this paper mainly includes two parts.Firstly,the Mengding Huangya,Zhu Yeqing and Ganlu tea are taken as the experimental subject,and the hyperspectral images are obtained through the hyperspectral image acquisition system.Image features and spectral characteristics are extracted from the preprocessed hyperspectral image,and 3 tea variety discriminative model(called model 1,model 2,model 3 respectively)based on spectral characteristics?image characteristics,and spectral-image fusion characteristics.The SPA algorithm is used to select feature and eliminate data redundancy.The experimental results show that using full-spectrum features combined with SVM(model 1)can discriminate 100% of different tea varieties;GLCM texture feature based on feature image combined with SVM model(model 2),the discriminative accuracy rate for different varieties of tea reached 100%.The SPA algorithm can select 10 feature variables from the original 72 image features and select 26 features from 319 spectral characteristics.Model 3 discusses the effect of different pre-processing methods for classification.It is determined that "Minmax" is the best pre-processing method and 100% of the classification accuracy is obtained in the test set.After determining the spectral information and image information of the hyperspectral image to be able to distinguish tea varieties,the feasibility of determining the tea grade was further explored.The discriminant model based on the spectral characteristics of the optimal internal quality grade of tea achieves 93.33% grade accuracy in the test set,but the accuracy of the best tea exterior quality grade discriminant model based on GLCM texture feature is only 86.67%.In order to fully exploit the texture information of the image dimension,the gray level co-occurrence matrix texture feature based on wavelet transform(WT-GLCM)was used as a new texture feature.The accuracy of the optimal tea grade discriminant model was 88.33%.In the study of tea grade discrimination using fusion features,three grade classification model called “original fusion characteristics-SVM”,“SPA screening characteristics-SVM”,“and PCA screening characteristics-SVM ” is compared.It is determined that PCA-based screening characteristics-SVM model is the optimal tea classification model,and achieves a classification accuracy of 96.67%,but the modeling time is only 0.77s.