Weld feature point detection is a key technology for welding trajectory planning and tracking. Existing two-stage detection methods and conventional convolutional neural network (CNN)-based approaches encounter performance bottlenecks under extreme welding noise conditions. To better obtain accurate weld feature point locations in high-noise environments, we propose a feature point detection network, YOLO-Weld, based on an improved You Only Look Once version 5 (YOLOv5). By introducing the reparameterized convolutional neural network (RepVGG) module, the network structure is optimized, enhancing detection speed. The utilization of a normalization-based attention module (NAM) in the network enhances the network’s perception of feature points. A lightweight decoupled head, RD-Head, is designed to improve classification and regression accuracy. Furthermore, a welding noise generation method is proposed, increasing the model’s robustness in extreme noise environments. Finally, the model is tested on a custom dataset of five weld types, demonstrating better performance than two-stage detection methods and conventional CNN approaches. The proposed model can accurately detect feature points in high-noise environments while meeting real-time welding requirements. In terms of the model’s performance, the average error of detecting feature points in images is 2.100 pixels, while the average error in the world coordinate system is 0.114 mm, sufficiently meeting the accuracy needs of various practical welding tasks.

To address the issues of intense image noise interference and computational intensity faced by traditional vision-based weld tracking systems, we propose WeldLight, a lightweight and noise-resistant convolutional neural network for precise classification and positioning of welding seam feature points using single-line structured light vision. Our approach includes (1) an online data augmentation method to enhance training samples and improve noise adaptability; (2) a one-stage lightweight network for simultaneous positioning and classification; and (3) an attention module to filter features corrupted by intense noise, thereby improving stability. Experiments show that WeldLight achieves an F1-score of 0.9668 for seam classification on an adjusted test set, with mean absolute positioning errors of 1.639 pixels and 1.736 pixels on low-noise and high-noise test sets, respectively. With an inference time of 29.32 ms on a CPU platform, it meets real-time seam tracking requirements.

Reliable weld seam perception remains challenging in industrial environments, where arc light, spatter, smoke, and varying seam geometries can seriously degrade visual sensing. These disturbances make it difficult to achieve a unified representation, accurate localization, and real-time inference at the same time. To address this problem, this paper presents an end-to-end lightweight framework for weld seam keypoint detection and tracking based on an improved SimCC. A unified three-keypoint formulation is introduced to represent different weld geometries by using one seam center point and two orientation reference points, thereby supporting a perception-to-control mapping in which position control and orientation control are decoupled. In addition, a lightweight C3k2-based backbone is designed, and a non-parametric log-domain quadratic peak-refinement decoder is proposed to alleviate the discretization-induced quantization error of SimCC classification distributions without adding model parameters. Experiments show that the proposed model contains only 1.4 M parameters, achieves 17.01 ms CPU inference latency, and obtains a detection accuracy of 1.89 px MAE. In curved weld seam tracking experiments with the integrated robotic system, it further achieves an average trajectory tracking error as low as 0.159 mm and an average orientation error of 3.738°, demonstrating its real-time accuracy and robustness for industrial welding applications.

To address the difficulty in stably detecting weld seam keypoints under complex industrial interferences, such as intense arc light, spatter, reflection, and partial occlusion, this paper proposes a lightweight weld seam keypoint detection model named StripePoint-YOLO. The proposed method formulates five typical types of weld seams as a unified detection-based keypoint localization task. Built upon YOLO11n, the model introduces a P2 detection head to enhance shallow high-resolution feature representation for keypoints and adopts SPDConv to reduce the loss of local details caused by early-stage downsampling. Meanwhile, the P5 detection output layer is removed, while its deep semantic features are retained for top-down feature fusion. This design reduces the negative influence of redundant coarse-scale predictions on the center localization of tiny keypoints. For optimization and training, WIoU v3 and NWDLoss are adopted as a joint regression loss to improve the stability of small-scale keypoint bounding box regression. In addition, an online physics-driven data augmentation strategy, OPDDA, is designed to simulate welding disturbances such as arc light, spatter, and dynamic occlusion. Experimental results show that StripePoint-YOLO achieves an mAP@50-95 of 80.46%, a Mean Center Error (MCE) of only 2.33 px, a parameter count of 1.86 M, and a computational cost of 19.42 GFLOPs, while reaching an inference speed of 159.80 FPS under the reported hardware configuration. Further MCE visualization and localization error analysis demonstrate that the proposed method maintains stable keypoint center localization across multiple weld seam types and complex noisy scenarios, verifying the effectiveness of StripePoint-YOLO for accurate and efficient weld seam keypoint detection in industrial welding images.