Hybrid ResNet–UNet method for multi-parameter acoustic hologram

Abstract

Acoustic holography has demonstrated broad application prospects in many fields. However, traditional computation methods for acoustic holograms still suffer from the challenges in accuracy, real-time performance, and stability under complex application scenarios. To address these issues, we propose a computational approach for parallel prediction of the amplitude and phase distributions on the hologram plane, using a deep feature fusion mechanism that integrates both residual network (ResNet) and UNet architectures. Simulation results demonstrate that on standardized test sets, the mean squared error reaches 0.0742 for phase prediction and 0.0085 for amplitude prediction, while reducing the computational time consumption by 8.46 times compared to conventional algorithms. The proposed ResUNet fusion method enables rapid and accurate prediction of both the amplitude and phase distributions on the hologram plane, which can then be propagated to the target plane to generate the desired acoustic field, offering a new technological pathway for advancing acoustic holography.