In-Situ Wavefront Correction via Physics-Informed Neural Network

Wavefront distortions pose a significant limitation in various optical applications, hindering further advancements in optical system performance. In this study, a novel generic calibration model based on Zernike-fitting neural network (ZFNN) is proposed, which enables insitu wavefront correction with just a single-shot measurement. The experimental setup follows a standard or equivalent focal-field imaging optical path, allowing calibration without the need to remove any components from the optical system. The ZFNN, a physics-informed neural network, offers the advantage of not requiring prior training, eliminating the need for extensive labeled data. With a fully connected network architecture and a modest number of neurons (469), the ZFNN achieves exceptionally fast optimization speed and meets the basic requirements for real-time calibration. Consequently, this approach holds great potential for applications such as rapid calibration of optical systems, high-precision light field modulation, and various advanced imaging techniques. This study presents an innovative Zernike-fitting neural network (ZFNN) for insitu wavefront correction, requiring just a single measurement without extensive pre-training or data. The proposed method demonstrates rapid calibration, streamlined optical path, and significant versatility has the advantages of fast calibration, simplicity of optical path and excellent versatility, which are confirmed by experimental validation, showing its potential in wavefront engineering applications. image