Deep-learning-augmented microscopy for super-resolution imaging of nanoparticles.

Area of Science:

• Nanotechnology
• Optical Microscopy
• Deep Learning

Background:

• Conventional optical microscopes struggle with subwavelength nanostructures, yielding blurry images.
• Diffraction-limited patterns contain hidden intensity and phase information crucial for nanoscale feature recognition.

Purpose of the Study:

• To develop a deep-learning framework for improving the spatial resolution of optical imaging for metal nanostructures.
• To enable super-resolution imaging of regularly arranged and randomly clustered nanoparticles and nanowires.

Main Methods:

• A convolutional neural network (CNN) was constructed and pre-trained using optical and scanning electron microscopy images.
• The CNN was trained to recover super-resolution images from blurry optical inputs of nanoparticles and silver nanowires.

Main Results:

• The CNN successfully recovered super-resolution images of nanoparticle dimers and multimers, accurately reconstructing profiles and orientations.
• The framework effectively deblurred images of cross-linked silver nanowires, with minor discrepancies at intersections.

Conclusions:

• The deep-learning augmented framework offers computational super-resolution optical microscopy.
• This approach has potential applications in bioimaging, nanoscale fabrication, characterization, and enhancing scanning electron microscopy resolution.