Application of Image Texture for Discrimination of Tea Categories Using Multi-spectral Imaging Technique and Support Vector Machine

Abstract

This study investigated multi-spectral imaging technique as a rapid method to discriminate the tea category. Tea was spread over the whole images. The images for each sample were captured using a red, near infrared and green channel multi-spectral camera. 320 images were obtained. Three texture features were obtained through the entropy of three channels and then set as the input variables for pattern recognition. Principal components analysis (PCA) and least squares-support vector machine (LS-SVM) were used for the pattern recognition. The cluster ability of PCA cluster plot was not good while the discrimination rate of LS-SVM model was 97.5%. We used one channel image to subtract another one, and six images of each sample were obtained. Then six new entropy values were obtained. The cluster ability of the new PCA cluster plot is better than the old one and the discrimination rate of LS-SVM model was 100%. It is concluded that multi-spectral imaging technique can identify categories of green tea fast and non-destructively.