EUVIMEDEUVIMED entdecken
Crossref

Poster Layout Optimization Driven by Deep Reinforcement Learning

Yiqiu Yu

Ingegneria Sismica · 30.04.2026 · 42 (4) · 1-22

DOI: 10.65102/is2026321 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Ingegneria Sismica, online erschienen am 30.04.2026. Verfasst von Yiqiu Yu.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Poster advertising layout optimization requires collaborative control of visual hierarchy, semantic primary and secondary, and spatial balance. However, manual design is costly and unstable across scenes. We propose a deep reinforcement learning framework that models layout generation as a sequential decision process of title, body, image, logo, price, button, and whitespace. The multi-branch state encoder integrates geometric properties, saliency cues, alignment relationships, and semantic importance, and the reward function jointly evaluates readability, balance, overlap penalty, information priority, and aesthetic consistency. Experimental results on 12480 poster samples show that the layout quality score of the model is 91.3, and the number of labeled elements is 74860. Compared with the rule template method, the accuracy of element alignment is improved from 84.7% to 92.6%, the overlap rate is reduced from 6.8% to 1.9%, and the average inference time is 0.18 s per sample. The framework provides a computable and scalable solution for intelligent layout optimization of posters in digital design systems for commercial scenes.

Zitieren

APA

Yu, Y. (2026). Poster Layout Optimization Driven by Deep Reinforcement Learning. Ingegneria Sismica, 42(4), 1-22. https://doi.org/10.65102/is2026321

Vancouver

Yu Y. Poster Layout Optimization Driven by Deep Reinforcement Learning. Ingegneria Sismica. 2026;42(4):1-22. doi:10.65102/is2026321

BibTeX
@article{Yu2026,
  author = {Yu, Yiqiu},
  title = {Poster Layout Optimization Driven by Deep Reinforcement Learning},
  journal = {Ingegneria Sismica},
  year = {2026},
  volume = {42},
  number = {4},
  pages = {1--22},
  doi = {10.65102/is2026321}
}
Kostenlos starten →