Deep Learning-Based Multi-frequency Phase Unwrapping for Single-Shot Composite Fringe Projection Profilometry

Fringe projection profilometry (FPP) is a widely recognized technique for three-dimensional (3D) surface measurement. However, conventional FPP often relies on multiple projections for phase unwrapping, which restricts measurement speed in applications such as additive manufacturing (AM) surface monitoring. In this work, we present a deep learning-based method that accurately retrieves phase information from a single-frame, frequency-multiplexed composite fringe. Specifically, three distinct horizontal fringe frequencies are encoded using separate vertical carrier frequencies and projected onto the test object. The captured composite fringe image is then processed by a trained neural network to calculate the wrapped phase corresponding to each embedded fringe pattern. The wrapped phase is then converted to the absolute phase using a multi-frequency phase unwrapping method. Ground truth phase data for training are obtained via high-accuracy phase-shifting profilometry. Experimental results demonstrate that the proposed approach achieves rapid and accurate unwrapped phase retrieval, enabling high-fidelity 3D surface reconstructions suitable for high-speed measurement scenarios.