In situ interpretation of complex physical processes in single trapped particles via deep-learning-driven solutions

We present deep-learning solutions aiming to elucidate complex physical processes underlying optical manipulation and luminescence emission during light-matter interactions at the single nanoparticle level. We illustrate two study cases considering aqueous and vacuum environments. Firstly, we employ holographic laser tweezers to control the trajectory, speed and rotation of single gold nanorods in an aqueous solution. We integrate polarization correlation spectroscopy of laser light scattered by a rotating nanorod with an unsupervised deep-learning framework to extract information about the particle’s geometry. Our neural networks also enable to correct aberrations in structured trapping beams. Secondly, we trap single upconversion particles in vacuum using a quadrupole Paul trap, whose luminescence spectra combined with supervised deep learning simultaneously provides the particle temperature and the impinging laser intensity. These AI-based solutions hold significant potential for advancing in-situ characterization of light-matter interactions at the nanoscale, enabling real-time, precise interpretation of particle dynamics, geometry, temperature, and illumination conditions.