Understanding radiative transport in porous media is essential for diverse engineering and industrial applications, particularly in the design of micro and nanoscale systems. However, the inherent complexity and non-linearity of porous packed beds, comprised of diverse particles and sizes, pose formidable challenges. To address these complexities efficiently, we develop a customized Physics-Informed Hybrid Deep Learning model, focusing on radiative characteristics due to light-matter interactions. Our investigation initiates with engineered geometric feature characterization. We demonstrate the significance of tailored directional geometric features (i.e., in-situ geometric features, and in-vitro geometric features). This comprehensive approach allows precise approximations of radiative properties while accommodating intricate structures and a multitude of dependentindependent variables at the micro-scaled porous medium. Our study extends to enhance optical radiative properties within nano-scale porous media (e.g., thin film solar cells). Nano-porous materials exhibit high surface area-to-volume ratios, promoting enhanced light absorption and scattering phenomena whereas engineered nano-pores can act as resonators, capturing light and inducing photonic band gaps that selectively filter wavelengths. To integrate these attributes into the process of learning and estimating radiative properties in a computationally cost-effective way, directional feature characterization and hybrid deep learning approach play a vital role. In summary, our research showcases the application of computationally feasible Physics-Informed Hybrid Deep Learning models to estimate radiative properties in micro and nanoscale porous medium. This novel approach enhances our understanding of the underlying physics and fosters emerging applications for polydisperse porous medium.