Super-resolution microscopy by grating and deep neural network

In this study, a novel optical super-resolution imaging technique, grating and deep neural network assisted super-resolution microscopy, is proposed. The technique utilizes a sub-wavelength grating, placed between the sample and the microscope objective, to convert the evanescent waves of a sample surface into propagating waves, allowing more high spatial-frequency information of the sample to be detected in the far field. Then, the far-field image of the sample is captured and trained end-to-end with a customized deep neural network model to heuristically reconstruct a clear image of the sample with structural features smaller than λ/3. Compared with the existing super-resolution imaging techniques, the proposed technique has the advantages of label-free, large field of view, one-time direct imaging, and white light illumination and observation in an atmospheric environment. Moreover, it has the flexibility to replace raster and network rendering components according to specific inspection requirements to meet diverse application scenarios.