Physics-Enhanced Multiscale Y-Net for Knife-Edge Wavefront Sensing

Wavefront error is a key metric for evaluating the quality of high-power laser components. During the early stages of optical component fabrication or in situ operation, rapid wavefront measurement is often required. Commonly used interferometric methods are difficult to meet this demand due to their complex optical setups and stringent environmental stability requirements. Therefore, there is a need for simpler and more efficient measurement approaches. The coherent knife-edge method is a convenient wavefront sensing technique with a simple optical configuration and fast acquisition speed, making it suitable for online detection. However, current inversion algorithms are limited in terms of accuracy and resolution. To address these limitations, this paper proposes Physics-Enhanced Multiscale Y-Net for Knife-Edge Wavefront Sensing. This method directly reconstructs the target wavefront from shadowgraphs, bypassing traditional numerical procedures. It improves both reconstruction accuracy and speed while enhancing detection resolution. The network takes two image-plane shadowgraphs as input and outputs the object wavefront. Training combines pre-training on simulated data with physics-informed fine-tuning. Simulation results show that the reconstructed wavefront closely matches the ground truth, with PV error within 6% and RMS error within 8%. Experimental results demonstrate PV error within 5% and RMS error within 8%. The approach demonstrates high accuracy across error metrics for wavefront detection, offering significant potential for efficient fabrication and inspection of high-power laser components.