Refinement of spectra using a deep neural network: Fully automated removal of noise and background

Abstract

We report the potential of U-Net deep neural network for the efficient removal of noise and background from raw Raman spectra. The U-Net method was first trained on simulated spectra and then tested with experimental spectra. The quality of the test results was quantified via different signal-to-noise ratios and the structural similarity index metric. The U-Net recovered Raman spectra feature a high structural similarity index, even for raw spectra that were dominated by background. The U-Net model does not rely on any human intervention.