Aiming at the low precision of welding seam appearance inspection, a welding seam appearance feature detection method based on optical low coherence was proposed. The three-dimensional coordinate information of the appearance of the weld was obtained by optical low coherence measurement principle, and the three-dimensional contour of the weld was extracted by filtering algorithm. Based on the characteristics of weld contour, DBGBCA algorithm was proposed to classify the point set to determine the fitting interval. After using RANSAC algorithm for linear fitting, the fitting interval and fitting order of polynomial are accurately adjustedfor curve fitting. Based on linear fitting and curve fitting, the characteristic points were determined, and the appearance parameters of the weld were obtained. Experimental results show that the accuracy of the proposed method can be up to 0.004 mm, which meets the requirements of industrial production.