High-resolution Holographic Image Reconstruction Based on Deep Learning

Aiming at the shortcomings of existing holographic reconstruction algorithms, which are complex and easily affected by noise, this project proposes a semantic partitioning U-Net for high-resolution reconstruction. Firstly, a method based on the edge-neural network is proposed to obtain more image semantic information and improve the model training effect. Secondly, the effective channel attention mechanism of deep neural networks is introduced to enhance the attention to the detailed information in the holographic image. This further improves the accuracy of the neural network. The convergence rate of the neural network is accelerated by introducing a linear element with leakage correction. Experimental results show that this method can quickly reconstruct phase and brightness images with better detail, better edge texture and flat background. Holographic images of various sizes can be reproduced. The research of this project will lay a foundation for applying holographic image enhancement technology based on deep learning.