Weld seam image feature recognition method with high robustness

The invention relates to the field of robot welding, in particular to a weld seam image feature recognition method with high robustness. A welding robot image feature recognition algorithm based on vision is designed, and a corresponding weld seam tracking visual model is established. A trajectory of a curve weld seam is determined through robot teaching, then pixels corresponding to the trajectory are mapped, the robot walks along the trajectory, and when weld seam deviation is found, weld seam deviation compensation is performed through visual servo control. Through the weld seam image feature recognition algorithm with high robustness and operation efficiency, path planning and deviation compensation can be quickly and accurately provided for the robot, and the welding quality and speedof the robot are effectively improved; and an image recognition system is not halted or crashed in the case of errors, disk failures, network overload or intentional attacks, meanwhile, through the method, a small computer m