Abstract: The electromagnetic scattering characteristics of atmospheric particles in the microwave and millimeter-wave frequency bands have a significant impact on radar system performance, precipitation identification, and propagation modeling. Traditional modeling methods often simplify particles as homogeneous structures, which fail to accurately capture their true inhomogeneous distributions, especially when the particle size is comparable to the wavelength. To improve modeling accuracy, this study adopts the Continuous Medium Approximation (CMA) to describe the internal structure of particles and employs the Invariant Imbedding T-Matrix (IITM) method for high-precision scattering calculations. As a recently developed electromagnetic computation technique, IITM offers broad applicability and good numerical stability, particularly suitable for modeling complex and inhomogeneous particles. However, when applied to continuous media, IITM still requires high-resolution 3D voxel discretization and suffers from frequent coordinate transformations and medium property lookups, resulting in high computational costs. To address this, a Fourier feature encoding multi-layer perceptron (Fourier-MLP) is proposed in this work to perform implicit modeling of binary continuous media structures. This method takes 3D spatial coordinates as input and rapidly predicts the corresponding medium properties, enabling efficient mapping between particle structure and the IITM computational domain. Experimental results demonstrate that the proposed method maintains physical consistency in modeling while significantly reducing the cost of scattering calculations, achieving high accuracy in both structural reconstruction and scattering response.