EUVIMEDEUVIMED entdecken
Crossref

SS-FDPNet: A Self-Supervised Denoising Network for Radiographic Images of Ship Welds

Xinhang Zhang, Yanjun Ma, Daofang Chang, Jiwang Du

IEEE Sensors Journal · 15.04.2025 · 25 (8) · 14235-14251

DOI: 10.1109/jsen.2025.3544236 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Sensors Journal, erschienen am 15.04.2025. Verfasst von Xinhang Zhang und 3 weiteren Autor:innen. Die Arbeit wurde bisher 5-mal zitiert (Crossref) und nennt 38 Literaturangaben. Gefördert von: National Key Research and Development Program of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The inspection of welding quality is a critical procedure to ensure the integrity and reliability of ship structures during construction. Radiographic nondestructive testing (NDT) is widely employed for its ability to deliver intuitive and precise inspection results without damaging the tested object. Deep learning approaches, particularly those utilizing convolutional neural networks (CNNs), have attracted considerable attention in the study of welding defect detection from radiographic images due to their superior image feature extraction capabilities. However, the inevitable noise present in such images significantly impairs the performance of these algorithms in target detection tasks. To address this issue, this article integrates the concept of image frequency decomposition and proposes an image denoising network model based on self-supervised learning (SSL). Initially, the characteristics of images across different frequency bands are analyzed, and spatiotemporal attention mechanisms are incorporated to design two types of frequency processing units and signal fusion units. Subsequently, leveraging expert knowledge, a self-supervised model is constructed to extract and mitigate noise characteristics. Finally, the proposed algorithm’s feature extraction performance is compared with several representative algorithms through the classification task of radiographic images. The results indicate that the convergence and recognition accuracy of the proposed model surpass those of other algorithms, achieving an overall ${F}1$ -score of 97.83%. In addition, the decision-making basis of the CNN model during the SSL process is visualized, enhancing the interpretability of CNN-based radiographic image recognition for welding inspection.

Förderung

  • National Key Research and Development Program of China

Zitieren

APA

Zhang, X., Ma, Y., Chang, D., & Du, J. (2025). SS-FDPNet: A Self-Supervised Denoising Network for Radiographic Images of Ship Welds. IEEE Sensors Journal, 25(8), 14235-14251. https://doi.org/10.1109/jsen.2025.3544236

Vancouver

Zhang X, Ma Y, Chang D, Du J. SS-FDPNet: A Self-Supervised Denoising Network for Radiographic Images of Ship Welds. IEEE Sensors Journal. 2025;25(8):14235-14251. doi:10.1109/jsen.2025.3544236

BibTeX
@article{Zhang2025,
  author = {Zhang, Xinhang and Ma, Yanjun and Chang, Daofang and Du, Jiwang},
  title = {SS-FDPNet: A Self-Supervised Denoising Network for Radiographic Images of Ship Welds},
  journal = {IEEE Sensors Journal},
  year = {2025},
  volume = {25},
  number = {8},
  pages = {14235--14251},
  doi = {10.1109/jsen.2025.3544236}
}
Kostenlos starten →