Abstract: Line-structured light vision sensors have been extensively applied in the field of robotic welding. Fillet welds composed of smooth and reflective welding plates made of stainless steel and aluminum alloy, the image characteristics of the structural light stripes at the weld can be contaminated by the surface reflective noise, so it is extremely difficult to identify and extract the characteristic points of the light strip weld. Aiming at this technical problem, an algorithm for identifying fillet weld seams resistant to surface reflection noise was proposed. Firstly, the algorithm groups and stores reflective noise stripes and structured light stripes, along with their respective positions, through row-by-row search during image preprocessing.Subsequently, by leveraging the stored light stripe positions and imaging characteristics of fillet weld light stripes, the algorithm distinguishes between glare noise and structured light stripe patterns on the grayscale image. Finally, it extracts the centers of structured light stripes and fits lines, and obtains characteristic points of the weld seams. Experimental results show the method has an accuracy of 92.4% in weld seam recognition, which can accurately identify the fillet welds contaminated by reflective noise. The method exhibits a higher recognition rate compared to certain comparative approaches.