Acoustic Holograms Dataset

To address the issue of insufficient real training data for acoustic holography, the Angular Spectrum Method (ASM) was used to synthesize an adequate amount of training data. This involved simulating a set of phase information on the holographic plane to provide ample training material for deep learning. During the training process, a custom learning rate scheduler was specifically employed to optimize the neural network's weight parameters, facilitating faster learning and enhancing network performance. This section also provides a detailed introduction to the deep learning approach of the IU-Net architecture, which is designed to more effectively process acoustic data and improve the quality of acoustic hologram reconstruction.