Abstract: Abstract: Corn production occupies an extremely important strategic position in grain production and grain security. Seeds are the most basic means of production in crop production, and the quality of seeds directly affects the subsequent crop cultivation, harvest yield and quality of grain. Moreover, directional seeding of corn can effectively improve the yield of corn, and it is necessary to select seeds before planting. Certainly, to choose good corn ears for threshing can get better corn seeds, which can lighten the workload of subsequent seeds selection and also improve efficiency. So, this study designed the scheme of corn ears selection device based on machine vision, and developed the selection algorithm for corn ears on the dynamic assembly line, which could complete the detection of the corn ears. First, the binary image of the corn ear to be detected was gained according to Otsu method, and after tracking the whole corn ear's contour, the length ratio between the long axis and short axis of the corn ear's contour was calculated. Then, by using the formula, R plus G minus B times 2, the area with yellow characteristics was strengthened and was even extracted after histogram threshold segmentation, and so the plumpness of the corn ear was detected by calculating the ratio of the extracted area and the whole area of the corn ear. Further, based on the characteristics of X cumulative distribution diagram, the middle ear row was extracted by tracking the seams' edge between ear rows. Also, based on the characteristics of ear row's Y cumulative distribution diagram, every seed in the middle ear row was extracted by using the method of threshold segmentation, and then the length-width ratio of their end-face was calculated by tracking their contour. And the flatness of the seeds was acquired. For the corn ears' video file acquired on the dynamic assembly line, only when the vertical coordinate of corn ear's center is located in image's central region, do the relevant frame image is tested. According to the characteristic parameter, the length ratio between the long axis and short axis of corn ears, the small corn ears were eliminated. The plumpness of corn ears, the corn ears which were mildewed or seriously lacked seeds were eliminated by the characteristic parameter, and, the flatness of corn ears, the corn ears in which the end-face of corn seeds were mostly round were eliminated by those parameters. This corn ears selection algorithm can rapidly and accurately detect the various characteristic parameters of corn ears, and finally determines their eligibility. In the experiment, 50 random samples of corn ears were detected, and the results showed that the determination accuracy of dynamic location was 100%, the recognition accuracy of identical corn ears was 100%, the detection accuracy of morphological characteristics was 100%, the detection accuracy of corn ears' plumpness (on the other hand, the degree of seed lack or mildew) was 96%, the detection accuracy of rectangular degree of seeds' end-face in corn ears is 98%, and the overall detection accuracy was 94%. This paper provides reference for the research on corn ears selection as well as corn seeds selection which is served for directional seeding. This paper provides reference for the research on corn ears selection early in corn seeds selection which is served for directional seeding.