Deep Learning-Driven Wavefront Sensing for Grating-Array-Based Wavefront Sensor

This article introduces a novel methodology for wavefront sensing through the integration of convolutional neural networks (CNNs) into grating array-based wavefront sensors (GAWS), providing a unique perspective on optical systems. In contrast to Shack-Hartmann wavefront sensors (SHWSs) that rely on lenslets, GAWS utilizes diffraction gratings displayed on a spatial light modulator (SLM) for focal spot array generation. Our focus is on replacing the traditional methods such as the Southwell algorithm with a CNN-based approach for wavefront estimation. The CNN model, trained on focal spot array images and corresponding Zernike modes, exhibits superior accuracy and efficiency compared to traditional methods. A comprehensive comparative analysis highlights the excellence of our CNN-based approach in wavefront reconstruction. The integration of CNNs into GAWS not only enhances wavefront sensing accuracy but also introduces a paradigm shift, exploring neural networks as viable alternatives to conventional methods.