Strong gravitational lensing has emerged as a powerful method for probing the nature of dark matter via substructure within galaxies. However, the limited availability of high-quality, high-resolution lensing images poses significant challenges to developing robust machine learning models, particularly for super-resolution imaging tasks. In this work, we present a novel, physics-informed approach to super-resolution of strong lensing images, designed specifically for sparse datasets. Unlike traditional supervised methods, our approach is fully unsupervised, requiring no ground-truth high-resolution images for training. By incorporating the gravitational lens equation directly into the architecture, our model is capable of extracting key physical information about the lens system, such as the distribution of dark matter in the lensing system, more efficiently in addition to enhancing image resolution. We validate our approach on simulated lensing datasets, demonstrating that our method not only improves image clarity but also provides meaningful insights into dark matter substructure. This work paves the way for more efficient analysis of upcoming large-scale surveys, including those from LSST and Euclid, which will dramatically expand the available data for strong lensing studies.