Advances in edge computing technology have greatly improved the real-time and accuracy of image recognition tasks. However, in order to efficiently migrate the image recognition task and achieve load balancing in the network to speed up the processing speed of the image recognition task, it is necessary to develop a migration algorithm. In this research, an in-depth analysis is conducted on the image recognition task migration scenario within the edge computing environment. The information regarding node load and computing power expectations is incorporated into the heuristic function of the ant colony algorithm, leading to the proposal of an innovative task migration algorithm. Simulation results show that the improved ACO algorithm reduces the task execution time by about 6% and 20%, and improves the load balance by about 35% and 67%, compared with the traditional ACO algorithm and the round robin algorithm, respectively.