<p>Driven by the growing global demand for high-quality teapots, regular quality inspections have become essential. Modern optical microscopes, significantly enhanced by recent technological advances, have become powerful tools in both materials research and medical science. In this study, we investigate Zisha teapots using optical microscopy. We first adopt polarization-induced microscopic imaging (PIMI) to obtain polarization-parameter images (I₀₀, sinδ, Φ) that reveal microstructural contrast beyond conventional intensity images. Then, we propose a tailored convolutional neural network (CNN) that operates on these polarization-parameter images to classify authentic versus counterfeit Zisha fragments, enabling a non-invasive evaluation of the pots’ material integrity. To ensure the physical reliability of the observed polarization contrast, the PIMI results were cross-validated using the Finite-Difference Time-Domain (FDTD) method, which simulates electromagnetic interactions with the material. Optical microscope images of the Zisha teapots are presented, achieving a super-resolution significantly higher than that of conventional imaging techniques. The integration of PIMI, deep learning, and FDTD provides a multi-layered approach, combining experimental imaging, data-driven classification, and computational physics to enhance inspection accuracy. Experimental outcomes demonstrate that the deep learning model achieves high classification accuracy, while FDTD validation confirms its alignment with the physical properties of traditional Zisha materials. This interdisciplinary methodology offers a novel solution for preserving and analyzing heritage artifacts, bridging the gap between advanced machine learning and classical material science techniques.</p>