Abstract: According to the morphological characteristics of rice，computer vision system was designed to capture rice dynamic image．In order to extract the image features ofthe rice in the moving state．the background substraction method was used to segment the current frame image and the background image．The color and morphological features were extracted from rice dynamic images．Using multi—structure neural network，the rice was graded into four classes：sound normal kernels，sound chalky kernels，cracked kernels and yellow kernels．Th e respective recognition accuracy of each classification was 95．2％ ，89．6％ ，97．3％ and 90．5％．The recognition results were satisfying and this method developed a reference for detecting rice image on line．

Key words: background subtraction, rice, dynamic image, multi—structure neural network, recognition