基 于 深 度 学 习 单 帧 干 涉 解 耦 的 平 行 平 板光 学 均 匀 性 高 精 度 测 量

Parallel plates, as typical transmission optical components, are widely employed in optical de⁃ tection, semiconductor manufacturing, and defense applications. Their optical homogeneity has a signifi⁃ cant impact on overall system performance. However, existing high-precision measurement methods face challenges, including difficulty in decoupling interference aliasing, time-consuming multi-frame acquisi⁃ tion, and limited capability in suppressing environmental noise. To address these limitations, a high-preci⁃ sion, deep learning-based single-frame interferometric decoupling method for measuring the optical homo⁃ geneity of parallel plates is proposed. First, a mapping model between aliased and single-sided interfero⁃ grams is constructed to decouple the single-frame aliased interferogram, enabling the retrieval of single-sid⁃ ed interferograms for the front and back surfaces and effective separation of interference fringes. Subse⁃ quently, a virtual phase-shifting reconstruction is performed to generate a sequence with equal phase-shift intervals from the single-frame interferogram. In combination with the conventional five-step phase-shift⁃ ing method, phase extraction and surface profile reconstruction are achieved, enabling high-precision evalu⁃ ation of optical homogeneity. A two-stage convolutional neural network is developed, in which the first stage performs the mapping from aliased fringes to single-sided fringes, and the second stage generates the five-step phase-shifting sequence and reconstructs the surface profiles of the front and back surfaces. In ad⁃ dition, a deep learning-based single-frame interferometric decoupling experimental system is established for optical homogeneity measurement. Experiments conducted on Φ75 mm and Φ50 mm parallel plate samples demonstrate that the proposed method yields results in good agreement with those obtained using a ZYGO interferometer, with absolute deviations on the order of 10^-7. The proposed approach enables high-precision and rapid measurement of optical homogeneity using only a single-frame aliased interfero⁃ gram, providing an effective solution for high-throughput and in situ inspection of optical components.