EUVIMEDEUVIMED entdecken
Crossref

Efficient model for detecting steel surface defects utilizing multi-scale feature fusion and multi-scale attention mechanism

Quan Lu, Minsheng Gong, Linfei Yin

Measurement Science and Technology · 10.04.2026 · 37 (15) · 155401

DOI: 10.1088/1361-6501/ae599e ↗

Kurz gefasst

Fachartikel in der Zeitschrift Measurement Science and Technology, online erschienen am 13.04.2026. Verfasst von Quan Lu und 2 weiteren Autor:innen. Die Arbeit nennt 49 Literaturangaben. Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In steel manufacturing, assessing surface defects is an essential quality control procedure that requires rapid and precise detection to maintain production efficiency. Current methodologies, however, encounter two primary obstacles: (1) the diminished contrast of defects within intricate backgrounds compromises detection precision, and (2) the high computational demand of feature extraction impedes real-time performance. To overcome these limitations, this paper introduces the multi-scale feature fusion and attention mechanism (MFAM)-YOLO, which incorporates three key strategies: (1) a lightweight block based on dynamic convolution and ghost module (DG-Block) for efficient feature extraction, balancing accuracy and speed; (2) the efficient slim generalized feature pyramid network is designed for facilitating the refinement and integration of feature information at different levels, thereby improving generalization capabilities and classification accuracy; (3) an efficient multi-scale attention mechanism before the detection head to improve feature aggregation and multi-scale target recognition in complex scenes. Empirical evaluations demonstrate that MFAM-YOLO attains an mAP of 81.0% on the NEU-DET benchmark with a throughput of 130 FPS, while yielding 75.1% mAP at 126 FPS on the GC10-DET dataset. These metrics indicate that the proposed approach achieves competitive accuracy and efficiency compared to existing methods.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Lu, Q., Gong, M., & Yin, L. (2026). Efficient model for detecting steel surface defects utilizing multi-scale feature fusion and multi-scale attention mechanism. Measurement Science and Technology, 37(15), 155401. https://doi.org/10.1088/1361-6501/ae599e

Vancouver

Lu Q, Gong M, Yin L. Efficient model for detecting steel surface defects utilizing multi-scale feature fusion and multi-scale attention mechanism. Measurement Science and Technology. 2026;37(15):155401. doi:10.1088/1361-6501/ae599e

BibTeX
@article{Lu2026,
  author = {Lu, Quan and Gong, Minsheng and Yin, Linfei},
  title = {Efficient model for detecting steel surface defects utilizing multi-scale feature fusion and multi-scale attention mechanism},
  journal = {Measurement Science and Technology},
  year = {2026},
  volume = {37},
  number = {15},
  pages = {155401},
  doi = {10.1088/1361-6501/ae599e}
}
Kostenlos starten →