Abstract
The inverse Mie scattering problem (IMSP) is extensively studied across various scientific fields due to its relevance in characterizing particles through light scattering. In infrared microspectroscopy, effects of Mie-like scattering significantly bias absorbance spectra complicating studies of microscopic objects. A general solution of the IMSP would allow to restore chemical information on a sample without any preknowledge about the absorption properties of the samples. Herein, we report on a deep-learning based Mie scattering correction that can be universally applied to an infrared spectrum of any sample kind. We compared the novel method with other approaches that were developed for and are valid only for specific types of samples. For validation we use a wide range of real-world validation samples such as microplastic beads, lung cells, and filamentous fungi that were measured in various spectroscopic setups, including a single detector and a focal plane array detector. Finally, we shed light on the uniqueness of the IMSP for spectral data. We find that in the vicinity of the true solution of the IMSP, all solutions are getting a characteristic distortion that cannot be typically observed in spectra, and therefore can be effectively sieved out by the suggested approach. The novel approach allows for the first time the retrieval of infrared spectra for infrared microspectroscopic studies in a quick way without requiring preknowledge about absorption properties of the samples investigated. Our approach offers a transferable framework for solving inverse Mie scattering problems across diverse scientific fields.