Vision monitor system is far from effective in the perception of the grainy images comprised of a large number of local homogeneous particles or fragmentations without obvious foreground and background. To overcome this problem, the sequential fragmentation theory is introduced into the visual analysis of the spatial structure of images. A kind of multi-scale and omni-directional Gaussian derivative filter is presented. The omni-directional structural features of the rice images under different observation scales are achieved by established the image Weibull distribution model. Auto- classification of the rice processing- quality is realized based on the principle of least squares-support vector machine. Abundant confirmative and comparative tests indicate the effectiveness and outperformance of the proposed method.