Uncertainty-aware physics-informed high-fidelity ghost imaging in ambient illumination

Ghost imaging excels in extreme conditions owing to its exquisite sensitivity and adaptability, yet ambient-light perturbations disrupt the masks and compromise reconstruction fidelity. We propose an uncertainty-aware physics-informed framework that explicitly characterizes target and mask distortions in ambient illumination. By integrating a dual-branch architecture with a progressive training strategy, image reconstruction is disentangled from noise suppression, enhancing stability and fidelity. Comprehensive verification demonstrates that the work alleviates reliance on a precise measurement matrix, achieving high-fidelity imaging in complex lighting and noise interference.