کنفرانس مهندسی برق ایران

صفحه اصلی / سی و سومین کنفرانس بین المللی مهندسی برق

GAN-Driven Image Generation for Metamaterial Absorbers Using Mean and Variance Encoding

نویسندگان :

Atefe Shahsavaripour¹ Mohammad Hossein Badiei² Leila Yousefi³ Ahmad Kalhor⁴

1- دانشگاه تهران 2- دانشگاه تهران 3- دانشگاه تهران 4- دانشگاه تهران

کلمات کلیدی :

Generative Adversarial Networks (GANs،Metamaterial Absorbers،Vari- ational Autoencoders (VAEs)،Deep Learning

چکیده :

In this work, we introduce an efficient methodology in the generation of complex images within a deep learning framework that combines Variational Autoencoders with Generative Adversarial Networks. In this work, the proposed methodology is applied in electromagnetics for the design of wideband metamaterial absorbers. This greatly simplifies the design process by encoding key image properties into compact low-dimensional representations, and each image can be represented with four numerical values that guide the GAN to generate high-quality outputs. This approach is very suitable in applications that demand efficient, accurate image generation, like in antenna design, electromagnetic imaging, and metamaterial structures. The present method will be used in the encoding of structural parameters within the latent space via deep VAE architecture while creating new feature vectors through the utilization of GANs in allowing creative and optimized designs concerning metamaterials. Simulated structures decoded and validated in CST Microwave Studio showed excellent absorption in a wide frequency range and for different angles of incidence. Such simplification of the design provides high-performance absorbing materials that could be used in critical applications of radar and telecommunications. Our results underline the transformative potential of machine learning, especially GANs, for solving complex electromagnetic design problems, opening new frontiers in material design and applications of neural networks.