Our framework is for a learned neural network that can rapidly generate accurate acoustic hologram. The framework contains its architecture, networks, loss functions, and datasets for training.

Autoencoder architecture with encoder(neural network) and decoder(simulation method, angular spectrum method).

Trains the encoder to make the decoder output identical to the encoder's input.

The decoder is also implemented using the Tensorflow library for gradient flow. It should be noted if someone wants to modify and try new things based on these codes.

This framework is introduced through publication in IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control.

Please cite the following if you utilized ours or modified it for your purpose.

M. H. Lee, H. M. Lew, S. Youn, T. Kim, and J. Y. Hwang, "Deep learning-based framework for fast and accurate acoustic hologram generation," in IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 2022, doi: 10.1109/TUFFC.2022.3219401.

Python 3.8 Tensorflow 2.5 numpy 1.21.5 matplotlib PIL scipy tqdm skimage

• Set the variables in Variables.py.

Excitation frequency, spatial step size (pixel size), total grid size, hologram pixel size(lateral_resol), transducer diameter, target plane distance, 
parameters of the algorithms (ex. learning rate, batch size, loss), parameters of the dataset

• Select one of the algorithms in main.py.

1. Diff-PAT for phase-only holograms
2. Iterative angular spectrum approach (IASA)
3. Ours: HU-Net

• Load target binary image

target_img=algorithm.load_target_img(target_path,plot=True)

• Retrieve phase-only hologram

retrieved_phase=algorithm.get_phase(target_img,get_computation_time=True)

• Reconstruct the target acoustic field by propagating the retrieved hologram

propagated_pressure=algorithm.propagate(retrieved_phase,expand_ratio)
-- expand_ratio: Increase spatial sampling for accurate simulation

• Visualize and assess the result

algorithm.visualize(target_img, retrieved_phase, propagated_pressure)
algorithm.assess(target_img, propagated_pressure,expand_ratio)

• Save the results if needed

algorithm.save_result_pressure(propagated_pressure,letter=img_name, save_path)
algorithm.save_matrix(retireved_phase,letter=img_name, save_path)