eduzhai > Applied Sciences > Engineering >

Real-time 3D Nanoscale Coherent Imaging via Physics-aware Deep Learning

  • Save

... pages left unread,continue reading

Document pages: 15 pages

Abstract: Phase retrieval, the problem of recovering lost phase information frommeasured intensity alone, is an inverse problem that is widely faced in variousimaging modalities ranging from astronomy to nanoscale imaging. The currentprocess of phase recovery is iterative in nature. As a result, the imageformation is time-consuming and computationally expensive, precluding real-timeimaging. Here, we use 3D nanoscale X-ray imaging as a representative example todevelop a deep learning model to address this phase retrieval problem. Weintroduce 3D-CDI-NN, a deep convolutional neural network and differentialprogramming framework trained to predict 3D structure and strain solely frominput 3D X-ray coherent scattering data. Our networks are designed to be "physics-aware " in multiple aspects; in that the physics of x-ray scatteringprocess is explicitly enforced in the training of the network, and the trainingdata are drawn from atomistic simulations that are representative of thephysics of the material. We further refine the neural network predictionthrough a physics-based optimization procedure to enable maximum accuracy atlowest computational cost. 3D-CDI-NN can invert a 3D coherent diffractionpattern to real-space structure and strain hundreds of times faster thantraditional iterative phase retrieval methods, with negligible loss inaccuracy. Our integrated machine learning and differential programming solutionto the phase retrieval problem is broadly applicable across inverse problems inother application areas.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×