High-precision wavefront reconstruction from a single interferogram via physical orthogonal decoupling and virtual phase-shifting.

Single-shot interferometric absolute phase recovery remains challenging under complex degradations. To address this highly ill-posed inverse problem, we propose a physics-aware orthogonal decoupling network (POD-Net). By leveraging orthogonal phasor decomposition, POD-Net transforms fringe demodulation into a well-posed linear decoupling of orthogonal component fields. A physical constraint engine, incorporating polar phase cosine and amplitude envelope conservation priors, guides network convergence to synthesize high-fidelity virtual phase-shifting sequences for accurate phase extraction. Evaluations on real degraded samples demonstrate a 26-ms single-frame inference speed and an absolute wavefront root-mean-square error of 0.00149λ. This method endows deep learning with strict physical interpretability while achieving highly competitive reconstruction accuracy.