Real-time sparse-sampled Ptychographic imaging through deep neural\n networks Article Swipe
YOU?
·
· 2020
· Open Access
·
· DOI: https://doi.org/10.48550/arxiv.2004.08247
· OA: W4287814828
Ptychography has rapidly grown in the fields of X-ray and electron imaging\nfor its unprecedented ability to achieve nano or atomic scale resolution while\nsimultaneously retrieving chemical or magnetic information from a sample. A\nptychographic reconstruction is achieved by means of solving a complex inverse\nproblem that imposes constraints both on the acquisition and on the analysis of\nthe data, which typically precludes real-time imaging due to computational cost\ninvolved in solving this inverse problem. In this work we propose PtychoNN, a\nnovel approach to solve the ptychography reconstruction problem based on deep\nconvolutional neural networks. We demonstrate how the proposed method can be\nused to predict real-space structure and phase at each scan point solely from\nthe corresponding far-field diffraction data. The presented results demonstrate\nhow PtychoNN can effectively be used on experimental data, being able to\ngenerate high quality reconstructions of a sample up to hundreds of times\nfaster than state-of-the-art ptychography reconstruction solutions once\ntrained. By surpassing the typical constraints of iterative model-based\nmethods, we can significantly relax the data acquisition sampling conditions\nand produce equally satisfactory reconstructions. Besides drastically\naccelerating acquisition and analysis, this capability can enable new imaging\nscenarios that were not possible before, in cases of dose sensitive, dynamic\nand extremely voluminous samples.\n
