Researchers have proposed and demonstrated a new computational imaging method by the incorporation of a physical model into a deep neural network, according to a recent research article published in the journal Light: Science and Applications.
Deep neural networks have been widely used in computational imaging, but most of them need a large amount of labeled data to train.
Depending on the network architecture and amount of data used for training, the network training process can take several hours or even several days, said the article.
In the article, the researchers from the Chinese Academy of Sciences, the University of Stuttgart and the Massachusetts Institute of Technology proposed a widely applicable physics-enhanced deep neural network, or PhysenNet for short, without the requirement of training data.
They demonstrated the applicability of PhysenNet in phase imaging. When a single diffraction image was fed into a PhysenNet model, the network weight and bias factors will be optimized through the interplay between the neural network and the physical model.
The new method opens up a new paradigm for the design and use of artificial intelligence in computational imaging. (Xinhua)
A research team led by Assoc. Prof. PAN An, Prof. YAO Baoli and MA Caiwen from Xi'an Institute of Optics and Precision Mechanics of the Chinese Academy of Sciences reported a colorization method based on color-transfer filtering FPM (CFFPM).
Computational imaging (CI) techniques exploit optical device and calculation algorithms to reconstruct object information. In a recent study, investigators from the Chinese Academy of Sciences described how they combined an untrained neural network and physics knowledge t...
86-10-68597521 (day)
86-10-68597289 (night)
86-10-68511095 (day)
86-10-68512458 (night)
cas_en@cas.cn
52 Sanlihe Rd., Xicheng District,
Beijing, China (100864)
Copyright © 2002 - Chinese Academy of Sciences
