Robust Hotspot Detection in Photonic Crystals Using a Hybrid Capsule–Transformer Deep Learning Model

Hassan Salarabadi, Mohammad Saber Iraji, Keivan Najafi and Dariush Salimi

Complexity, 2025, vol. 2025, 1-16

Abstract: This study introduces a novel hybrid deep learning framework to enhance hotspot detection in photonic crystal design, addressing challenges in accuracy and overfitting. By integrating Capsule Networks (CapsNet) and transformer architectures with ensemble learning and data augmentation, the proposed approach optimizes the identification of wavelength propagation discrepancies in photonic structures. Experiments conducted on the ICCAD-2012 benchmark dataset demonstrate that the ensemble model achieves 90% test accuracy, outperforming standalone CapsNet and transformer models while reducing overfitting. The model also demonstrates strong classification consistency, with an F1-score of 90% and a G-mean of 90%, indicating robust performance across precision–recall balance and class-wise sensitivity–specificity harmony. The framework’s success in balancing performance and generalization highlights its potential to streamline photonic device design for applications in sensing, telecommunications, and energy harvesting. This work bridges advanced machine learning techniques with photonic engineering, offering a scalable and efficient solution for complex light-matter interaction analysis.

Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
http://downloads.hindawi.com/journals/complexity/2025/8211411.pdf (application/pdf)
http://downloads.hindawi.com/journals/complexity/2025/8211411.xml (application/xml)

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:hin:complx:8211411

DOI: 10.1155/cplx/8211411

Access Statistics for this article

More articles in Complexity from Hindawi
Bibliographic data for series maintained by Mohamed Abdelhakeem ().