Deep image restoration for infrared photothermal heterodyne imaging

Infrared photothermal heterodyne imaging (IR-PHI) is an all-optical table top approach that enables super-resolution mid-infrared microscopy and spectroscopy. The underlying principle behind IR-PHI is the detection of photothermal changes to specimens induced by their absorption of infrared radiation. Because detection of resulting refractive index and scattering cross section changes is done using a visible (probe) laser, IR-PHI exhibits a spatial resolution of ∼300 nm. This is significantly below the mid-infrared diffraction limit and is unlike conventional infrared absorption microscopy where spatial resolution is of order ∼5μm. Despite having achieved mid-infrared super-resolution, IR-PHI's spatial resolution is ultimately limited by the visible probe laser's diffraction limit. This hinders immediate application to studying samples residing in spatially congested environments. To circumvent this, we demonstrate further enhancements to IR-PHI's spatial resolution using a deep learning network that addresses the Abbe diffraction limit as well as background artifacts, introduced by experimental raster scanning. What results is a twofold improvement in feature resolution from 300 to ∼150 nm.