，选出率为80.7%。A cotton seed on-line separation system based on computer vision was designed to realize the automatic separation of red-brown and black cotton seeds produced in Xinjiang Province. In the system, the seeds were projected horizontally, separated by compressed airflow in the air. Meanwhile, a method of capturing seed images was studied in an unordered arrangement. And the corresponding relationship between unordered seeds and the airflow nozzles was resolved by region subdivision of seed image. Lastly, a seed tracking and separating algorithm was proposed which memorized the position of each inspected seed until the seeds reached the separation region. The experiment results show that when the speed of the conveyer belt is 0.50 m/s, the inspecting accuracy is 88.6%, and the separating ratio is 80.7%.