Abstract: Aiming at the problems of low efficiency and accuracy for roll coating thickness detection, a machine vision detection method based on heuristic ant colony algorithm was proposed. The principle of Canny operator was used to extract the edge information and obtain the prior knowledge of edge points. Then an improved edge tracking model of ant colony algorithm was established, which realized the guidance of pheromone and heuristic information to ants, avoided the distribution and walking of ants in non-edge areas and solved the problem of coordinated randomicity and positive feedback in traditional ant colony algorithm. This improved ant colony algorithm based on machine vision was used in the measurement experiment. Compared with the mechanism modeling method, the maximum error is 5.74%, and the average error is 4.04%, which meet the actual production needs.