Abstract:A method for classification of Xi-hu-long-jing tea in four grades was introduced based on machine vision of multi-spectral imaging technique. Firstly, three monochrome images at 540,670 and 800 nm wavelengths were simultaneously obtained based on 3CCD multi-spectral camera, then image features including 18 shape features and 15 texture features were extracted based on image processing technology. These two groups of features were adopted for cluster analysis with principal component analysis of the four grades tea. However the result was not satisfactory. In order to obtain a more effective separation model, the two groups of features were combined, and the cluster analysis was conducted again based on the combined features. The result was better than the former. After optimization of these three groups of features, three classification models were developed by means of multiple stepwise discriminant analysis (MSDA). It was found that model based on the combined features had the best performance with accuracy of 85% for prediction of unknown samples. The most important two features for classification were correlation and energy of 800 nm wavelength monochrome image.