Deep-learning-based denoising approach to enhance Raman spectroscopy in mass-produced graphene

The inherently weak signal present in Raman spectroscopy makes spectral resolution susceptible to noise. Hence, efficient denoising techniques for post-processing of spectral data are required. We introduce two efficient approaches to remove noise from graphene Raman spectra, based on deep neural network architectures using supervised and unsupervised learning. We compared the performance of these approaches with three traditional noise removal methods. The experimental results demonstrate the effectiveness of deep-learning models in the denoising task, which is crucial in interpreting characterization data of mass-produced graphene. Overall, our supervised approach outperforms all considered baselines, as well as the unsupervised method, providing significant improvement in noise reduction.