The Classification Of Tea And Stem Based On Statistical Shape Features

At present, the sorting of tea and stems, processed by automated sorting machine, has a low efficiency in China. Firstly, we create tea and stems images library, and then define and choose their available color and shape features by the techniques of digital image processing. Finally, we choose appropriate classifier to sort by the pattern recognition technology and achieved good results.In the part of image preprocessing, we firstly translate the RGB true color images to grayscale images and binary images, because the RGB images contain large volumes of data and other issues. It uses the method of weighted mean to change RGB images to gray images. Changing gray images to binary images, we use the background subtraction-threshold value iteration method. The binary images still have salt and pepper noise, so this paper uses mathematical morphology to delete them. The mathematical morphology processing contains opening operation and closing operation. Finally, the binary image library, used by features selection, is established.In the part of features extraction, we extract the features based on color and shape. According to the obvious difference between their shape, the area of differential mean is found and extracted in this paper.In the part of feature selection, this paper has three steps. Firstly, some features are directly eliminated, as they are apparent inseparable. Secondly, the correlation coefficient matrix is used to choose six features from them. Finally, we use the K-W method to select three best features. They are the second-order moment, the minimum inscribed circle and the area of differential mean.In the experimental part, the SVM and the minimum-risk Bayesian classifier are selected. The SVM has a high recognition rate and remain stability. The minimum-risk Bayesian classifier has a risk factor which can be adjusted according to the need. Through the analysis of experimental results as well as the comparison to other classifiers, the two classifiers have good classification results.