Speckle beam sensing via quadriwave lateral shearing interferometry assisted by convolutional neural networks

Quadriwave lateral shearing interferometry (QLSI) is a powerful quantitative phase imaging technique, offering compactness, robustness, high resolution, and single-shot measurement. However, since QLSI phase retrieval relies primarily on phase gradient integration, it may struggle with speckle beams containing phase singularities, where phase discontinuities lead to non-conservative gradients and reduce integration accuracy. Here, we propose a convolutional neural network (CNN)-assisted QLSI sensing method for speckle beam reconstruction. The CNN is trained to directly retrieve the complex amplitude information of speckle beams from their QLSI interferograms. Experimental results demonstrate that the proposed method can accurately reconstruct both amplitude and phase profiles of speckle beams with various complexities, achieving a Pearson correlation coefficient exceeding 0.9 compared to the ground truth. This method is expected to enhance the sensing capability of traditional QLSI wavefront sensors and enable broader applications, such as free-space laser communication and advanced optical imaging in strong turbulence scenarios.