Abstract: The key methods of tea image processing were studied, including threshold transforming, median filter, image mark and boundary following. These methods solved the problem of how to accurately extract and calculate the characteristic parameters of tea shape. The software for tea quality detection based on digital image was developed. The functions, such as collection of tea image, image processing and extraction of characteristic parameter, could be accomplished. Using the software, 17 shape characteristic parameters were collected from 108 pieces of three different kinds of tea. And 6 characteristic parameters were used to build the back propagation-artificial neural network (BP-ANN) model. The variety from thirty unknown samples were predicted by this model and the recognition rate of eighty percent was achieved.

Key words: tea, image processing, characteristic parameter, BP neural network