Recognition of weld seam for tailored blank laser welding based on least square support vector machine

Accurate recognition of weld seam was the key for structural-light visual inspection of weld quality with high precision in tailored blank laser welding. Because of the problem that when the distortion of laser stripe was not obvious in the image and the welding seam cannot be recognized accurately, a recognition method according to the texture information of weld seam based on least squares support vector machine was proposed in this paper. Firstly, the textural features of the image were analysed and the textural features which had obvious difference between weld seam region and non-welded region were extracted. Secondly, the least square support vector machine model was trained and the coarse recognition of weld seam was accomplished. Finally, a fine recognition was achieved by Laws texture filter and threshold segmentation. The recognition experiments were carried out for weld seam in different welding parameters and the results showed that the weld seam can be recognized effectively by this method.