Aiming at the problem of low segmentation accuracy of weld laser stripes caused by a large amount of arc light noise in the actual welding environment, A method for weld seam feature point recognition and localization integrating deep learning with digital image processing algorithms is proposed. Firstly, an image acquisition system designed for actual welding environments is developed. Secondly, a deep learning-based object detection model was trained to identify the weld seam region of interest (ROI), enabling real-time coarse localization of the laser stripe region. Thirdly, improved digital image processing algorithms is applied to the weld seam region of interest (ROI) to extract the pixel coordinates of feature point. Finally, the world coordinates of weld seam feature point is reconstructed using Three-dimensional reconstruction algorithm and the superiority and applicability of the algorithm is experimentally validated. Experimental results demonstrate that the feature point localization error for a single image remains within ±1 mm, with a mean processing time less than 0.5s. Even under complex welding environments containing interference such as arc glare and smoke noise, the proposed method maintains rapid and accurate detection of weld seam feature point.