EUVIMEDEUVIMED entdecken
Crossref

A lightweight pyramid-pooling and spatial-attention Unet for laser stripe segmentation in laser vision seam tracking

Junkai Hong, Liuyang Zhong, Jianzhao Wu, Fan Fei, Wenhui Liu, Chaoyong Zhang, Xiao Liu

Physica Scripta · 24.07.2026 · 101 (29) · 296001

DOI: 10.1088/1402-4896/ae87e7 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Physica Scripta, online erschienen am 17.07.2026. Verfasst von Junkai Hong und 6 weiteren Autor:innen. Die Arbeit nennt 42 Literaturangaben. Gefördert von: National Science and Technology Major Project, National Natural Science Foundation of China, Fujian Province Young and Middle-aged Teacher Education Research Project (Science and Technology) Key Project.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Weld tracking systems based on structured light vision have been widely used in robotic welding. The precise extraction of laser stripes is key to achieving automatic weld seam tracking based on laser vision. However, traditional image-based laser stripe extraction methods are often susceptible to welding-induced interference, including smoke, spatter, and arc light, which greatly limits their practical applications in industry. To solve this problem, this paper proposes a U-shaped laser stripe segmentation network, named CP-PS-Unet, which combines the copy-paste method, a lightweight pyramid pooling module, and a spatial attention mechanism. Furthermore, an image pre-processing method with addition of Gaussian noise, salt-pepper noise, random noise is also employed. The proposed CP-PS-Unet can effectively extract laser stripes from seam images captured by a laser structured-light vision sensor, realize the adaptive extraction of laser stripe features, and improve the real-time performance and robustness of the system. According to the experimental results, the Mean Intersection over Union, Kappa coefficient, and dice coefficient of the proposed CP-PS-Unet algorithm are 86.01%, 83.79% and 91.90%, respectively, representing improvements of 7.81%, 11.51% and 5.76% compared to the traditional Unet, and the accuracy is increased to 99.71%. Overall, the proposed CP-PS-Unet shows good segmentation performance and high robustness to seam images affected by strong welding interference. It can be used for the automatic extraction of laser stripes and provides an effective method for real-time laser vision seam tracking in robotic welding.

Förderung

  • National Science and Technology Major Project
  • National Natural Science Foundation of China
  • Fujian Province Young and Middle-aged Teacher Education Research Project (Science and Technology) Key Project
  • Municipal Natural Science Foundation of Xiamen
  • Ministry of Industry and Information Technology High-quality Development Special Project

Zitieren

APA

Hong, J., Zhong, L., Wu, J., Fei, F., Liu, W., Zhang, C., & Liu, X. (2026). A lightweight pyramid-pooling and spatial-attention Unet for laser stripe segmentation in laser vision seam tracking. Physica Scripta, 101(29), 296001. https://doi.org/10.1088/1402-4896/ae87e7

Vancouver

Hong J, Zhong L, Wu J, Fei F, Liu W, Zhang C, et al. A lightweight pyramid-pooling and spatial-attention Unet for laser stripe segmentation in laser vision seam tracking. Physica Scripta. 2026;101(29):296001. doi:10.1088/1402-4896/ae87e7

BibTeX
@article{Hong2026,
  author = {Hong, Junkai and Zhong, Liuyang and Wu, Jianzhao and Fei, Fan and Liu, Wenhui and Zhang, Chaoyong and Liu, Xiao},
  title = {A lightweight pyramid-pooling and spatial-attention Unet for laser stripe segmentation in laser vision seam tracking},
  journal = {Physica Scripta},
  year = {2026},
  volume = {101},
  number = {29},
  pages = {296001},
  doi = {10.1088/1402-4896/ae87e7}
}
Kostenlos starten →