An Integrated IGABEM-DNN-CGAN Framework for Efficient Electromagnetic Scattering Analysis of Two Dimensional Dielectric Structures

• A novel hybrid framework integrating IGABEM with a hierarchical DNN-CGAN surrogate is proposed for highly efficient and accurate electromagnetic scattering analysis. • Seamless NURBS-based integration of CAD and analysis via IGABEM eliminates geometric errors and ensures high-fidelity modeling of complex scatterers. • A two-stage hierarchical learning strategy employs a DNN for global mapping and a CGAN for refining fine-scale features, dramatically accelerating predictions while preserving physical consistency. • Superior performance is demonstrated for both canonical and complex geometries, showing excellent agreement with analytical solutions and IGABEM results under TE and TM polarizations. This study proposes a computational framework combining the Isogeometric Boundary Element Method (IGABEM) with deep learning models to analyze electromagnetic scattering of two-dimensional dielectric objects under TE and TM polarizations. By leveraging Non-Uniform Rational B-Splines (NURBS), the framework provides a unified representation of geometry and analysis, enabling precise modeling of complex boundaries. A hierarchical Deep Neural Network–Conditional Generative Adversarial Network (DNN–CGAN) surrogate is employed: the IGABEM dataset first trains the DNN, whose outputs are subsequently used to train the CGAN. This two-stage strategy markedly accelerates computation while maintaining high predictive accuracy. Numerical examples demonstrate the effectiveness and robustness of the proposed approach, confirming its suitability for rapid and reliable electromagnetic scattering analysis in complex engineering scenarios.