An improved YOLO11 network integrated with sample generation for the intelligent detection of steel plate surface defects

Luya Yang, Min Zhang and Yaxian Gao

PLOS ONE, 2026, vol. 21, issue 8, 1-29

Abstract: To satisfy the stringent surface quality requirements imposed for industrial steel plates, automated defect detection techniques must balance high accuracy with low computational latency. In this paper, a lightweight detection method integrating adaptive image enhancement, generative sample synthesis, and an optimized detection model is proposed. First, an AW-ACE algorithm is introduced to resolve low contrast levels by dynamically merging color channels using information entropy. Second, an ES-DCGAN model based on ECA and an SELU is used to synthesize diverse and high-quality defect samples. Finally, we design a lightweight model, i.e., GE-YOLO11n, by incorporating GhostConv and ECA into YOLO11n to optimize the feature extraction process for small-scale defects. The entire framework yields a 1.5% mAP improvement and a 0.3 ms detector-only latency reduction relative to the baseline. Ablation studies demonstrate that the complete pipeline, comprising the ES-DCGAN, AW-ACE, and GE-YOLO11n modules, improves the overall mAP by 6.3% and reduces the number of required parameters by 21.4%. Compared with the evaluated models, our proposed method achieves an mAP value of 91.3% and a full-pipeline latency level of 7.4 ms, outperforming the other compared models under the experimental conditions specified in this study. This method delivers a highly efficient and robust solution for attaining real-time industrial quality control.

Date: 2026
References: Add references at CitEc
Citations:

Downloads: (external link)
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0356877 (text/html)
https://journals.plos.org/plosone/article/file?id= ... 56877&type=printable (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:plo:pone00:0356877

DOI: 10.1371/journal.pone.0356877

Access Statistics for this article

More articles in PLOS ONE from Public Library of Science
Bibliographic data for series maintained by plosone ().