Hybrid neural network-based adaptive computational ghost imaging

We propose a hybrid neural network-based adaptive computational ghost imaging (CGI) method to restore clear images of objects with different sub-Nyquist sampling ratios (SNSRs). We design an adaptive method and a hybrid neural network for CGI-based image reconstruction. We use an interference-adding layer in the network of the proposed method (HA) to remove multiple degradations and noise during the training process. We train the network once with simulated data, and it can recover high-quality images from the experimental data at different SNSRs. The effectiveness and advantages of HA are numerically and experimentally studied. HA is helpful for improving the image quality and reconstruction efficiency of CGI.

Automated summary

The study introduces a hybrid neural network-based adaptive computational ghost imaging (CGI) method, which can restore clear images of objects with different sub-Nyquist sampling ratios. By utilizing an interference-adding layer in the network, the method effectively removes multiple degradations and noise during training, improving the efficiency and quality of image reconstruction.