Deep-learning-enhanced deflectometry for rapid surface measurement

In phase measuring deflectometry (PMD), achieving high-precision three-dimensional (3D) specular surface measurement often suffers from time-consuming issues. We propose a deep learning-enhanced deflectometry method for rapid and accurate 3D surface reconstruction. The proposed method integrates an end-to-end neural network with a zonal wavefront reconstruction algorithm, forming a hybrid framework that combines the advantages of deep learning in fast prediction and traditional iterative algorithms in high-precision refinement. Specifically, an end-to-end network first predicts initial surface profiles directly from bi-directional fringe patterns, and the results are then utilized as initial values for the iterative optimization algorithm. Simulation and experimental comparisons with different methods demonstrate that the proposed method achieves a remarkable reduction in initial value generation time and faster convergence speed. This hybrid framework combines the advantages of deep learning in fast prediction and traditional iterative algorithms in high-precision refinement, offering a new solution for efficient specular surface 3D reconstruction.