Features Extraction Of Fresh Tea Images And Its Application On The Recognition Of Tea Varieties

The cultivation and varieties research of tea are facilitated to the protection of tea elite cultivars. Based on fresh tea images in different varieties, this paper proposed a new method for the rapid recognition of tea varieties by systematic study about the acquisition and preprocessing of images, extraction and selection of feature, and multiple classifiers for classification.1. The acquisition process of image samples in10tea varieties was introduced in detail. In order to preprocess the original target images, B channel graying, Otsu background division and mathematical morphological processing were also used to erase the information outside the specified regions.2. Feature extraction methods on the color, shape and textural features were studied, and for the first time multi-fractal was applied to the recognition of fresh tea images. Four kinds of features, including color, shape, texture and fractal were extracted, all of which summed up to46features. Through the single factor variance analysis and correlation analysis of all the features, it was found the selected features have a significant impact on the tea category, but there are correlations between features.27features were screened by multi-roundly worst descriptor elimination methods.3.Based on each category and all the combinations of four kinds of features, SVMKM, Bayes, random forests, Fisher, K-nearest neighbor rule and extreme learning machine were used to build models, and their classification accuracies were compared. The results showed that, in10-fold cross validation, SVMKM and random forests respectively combined with all the27features hold the best independent prediction accuracy91%. In all, the method proposed in this paper was able to can effectively identify the varieties of tea.