The project aims to improve image compressive sensing (CS) reconstruction by developing deep unfolding networks that combine the interpretability of iterative CS algorithms with the speed of deep learning. Instead of tying each network block to a single iteration, the proposed approach treats each module as an independent optimization problem with its own CS model and penalty, designed to be more robust and less prone to local minima while enabling fast, high-accuracy reconstruction. The work also includes refining the underlying mathematical models, exploring multi-modal data integration, building a rigorous validation framework on synthetic and real datasets under varying noise/compression, improving scalability for large-scale/real-time use, and enhancing model interpretability for practical deployment.
