Phase retrieval method for single-frame point diffraction interferogram images based on deep learning

To address the issue of decreased measurement accuracy caused by environmental errors introduced by multi-step phase shifting in traditional point diffraction interferometry, a deep-learning-based phase retrieval method for single-frame point diffraction interferograms is proposed. Two neural networks, designed for different stages of interference fringe image processing, are constructed specifically for phase retrieval of point diffraction interferograms. A diverse dataset of point diffraction images is developed for training and optimization, enabling accurate and rapid processing to achieve high-precision phase unwrapping. The accuracy of this method is validated using actual images collected from a point diffraction experimental platform, and the results are compared with those obtained using ESDI professional interferogram processing software and other algorithms. The comparison demonstrates that the results are largely consistent, indicating that the proposed method is both fast and highly accurate in phase retrieval. This method provides a feasible solution for high-precision image processing in point diffraction interferogram analysis.