In order to solve the problems of low accuracy of manual counting and long time-consuming photoelectric methods such as grain counting instrument in the process of paddy grain detection, the computer vision system and MATLAB software development platform are used instead of manual paddy grain detection for automatic detection. The algorithm uses spatial filtering of rice grain images to remove noise “Otsu” method for optimal global threshold segmentation of the image, based on the image processing to complete the determination of the number of rice grains, grain size of the two indicators. Four varieties of rice, namely, ‘Wuchang Rice’, ‘Indica Rice’, ‘Japonica Rice’ and ‘Founder Rice’, were used as experimental subjects. Object, to study the determination of the two indexes of rice grain counting and rice grain size grading. The experimental results show that the algorithm and the program are accurate and effective, can accurately calculate the number of rice grains, and the algorithm and the program can accurately and effectively determine the size of rice grains. The algorithm and program to determine the average area size of each variety of rice particles and 100-grain weight into a positive correlation, and the coefficient of determination is 0.91965, the algorithm and program can accurately and effectively determine the size of soybean particles. In short, based on the research of MATLAB image processing rice grain detection method can relatively reduce the labor intensity of human labor and the lack of human vision, in improving the efficiency and accuracy and other aspects of the work of the important significance.