A seed point growing algorithm for ultra-narrow gap weld seam tracking with anti-interference capability

Ultra-narrow gap welding (0.05–0.20 mm) offers advantages in low heat input, reduced deformation, and improved material utilization, but its seam tracking remains highly challenging because the target seam is often near the image-resolution limit and embedded in strong industrial interference. Under such conditions, ultra-narrow gap seam tracking is essentially a weak-feature, strong-interference, and continuity-sensitive detection problem, in which true seam cues are incomplete, locally ambiguous, and easily confused with seam-like artifacts. To address this difficulty, this paper proposes a seed-point growing algorithm that emphasizes reliable seed establishment before seam growing. The method consists of three stages: reliable seed establishment, constrained seam growing, and row-based adaptive trimming and fitting. In the first stage, global multi-row interval sampling, iterative trimming, position clustering, and trial-growing verification are combined to identify a reliable seed point under weak and ambiguous seam indications. Based on this seed, constrained growing is performed using customized local window adaptive thresholding, direction continuity constraints, and cyclic breakpoint tracking to recover continuous seam features under coupled interference. Finally, median position constraints and quadratic polynomial optimization are integrated to generate a smooth localization trajectory. Experiments on SiC/Al composite, aluminum alloy, and pure copper specimens under joint-like thin-line, high-reflection, complex-marking, and composite interference conditions show that the proposed method achieves a positioning error of approximately 1 pixel (0.063 mm) with a processing time of 0.18–0.65 s. These results demonstrate stable anti-interference performance and indicate the method's potential for practical ultra-narrow gap weld seam tracking in industrial applications.