EUVIMEDEUVIMED entdecken
Crossref

ProDAS: Prompt-Driven Adaptive Segmenter for Surface Defect Segmentation

Zhiyong He, Chuanning Wang, Guo Yi, Yanzhao Zhou, Zhenxiong Gu, Song Lin

IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 5013116-5013116

DOI: 10.1109/tim.2026.3706200 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Zhiyong He und 5 weiteren Autor:innen. Die Arbeit nennt 49 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Open Research Fund of Anhui Province Key Laboratory of Machine Vision Detection and Perception..

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Precise surface defect segmentation is vital for industrial quality control, yet the performance of deep learning models is often constrained by the availability of large-scale annotated data. This challenge stems from the prevalent implicit learning paradigm, which limits the models’ ability to generalize from sparse data where high-level semantic context is crucial. To address this, we propose a prompt-driven adaptive segmenter (ProDAS), a novel framework built on a synergistic learning paradigm that augments a data-driven visual backbone with explicit knowledge from natural language prompts. ProDAS employs a dual-guidance mechanism: a vision-language attentive masker (VLAM) translates prompts into strong spatial priors, while a factorized semantic alignment (FSA) module injects this textual guidance at multiple scales into the backbone. The visual backbone is further enhanced by a multiattention feature refinement module (MAFRM) to capture discriminative features. Comprehensive evaluations demonstrate state-of-the-art performance, with the framework excelling in data-scarce scenarios. For instance, it achieves an impressive 80.6% mIoU on short weft (SW) defects in a one-shot setting. Its high robustness against industrial degradations, coupled with an inference speed of 49 FPS on an NVIDIA RTX 4060 8G GPU, validates the efficacy of our synergistic paradigm and establishes ProDAS as a practical solution for real-world manufacturing. The source code is publicly available at https://github.com/LVLVN/prodas

Förderung

  • National Natural Science Foundation of China
  • Open Research Fund of Anhui Province Key Laboratory of Machine Vision Detection and Perception.

Zitieren

APA

He, Z., Wang, C., Yi, G., Zhou, Y., Gu, Z., & Lin, S. (2026). ProDAS: Prompt-Driven Adaptive Segmenter for Surface Defect Segmentation. IEEE Transactions on Instrumentation and Measurement, 75, 5013116-5013116. https://doi.org/10.1109/tim.2026.3706200

Vancouver

He Z, Wang C, Yi G, Zhou Y, Gu Z, Lin S. ProDAS: Prompt-Driven Adaptive Segmenter for Surface Defect Segmentation. IEEE Transactions on Instrumentation and Measurement. 2026;75:5013116-5013116. doi:10.1109/tim.2026.3706200

BibTeX
@article{He2026,
  author = {He, Zhiyong and Wang, Chuanning and Yi, Guo and Zhou, Yanzhao and Gu, Zhenxiong and Lin, Song},
  title = {ProDAS: Prompt-Driven Adaptive Segmenter for Surface Defect Segmentation},
  journal = {IEEE Transactions on Instrumentation and Measurement},
  year = {2026},
  volume = {75},
  pages = {5013116--5013116},
  doi = {10.1109/tim.2026.3706200}
}
Kostenlos starten →