The article has been published online, in open access, by Advanced Engineering Materials (Wiley). It presents a practical Noise2Noise-based denoising pipeline for high-throughput Raman spectroscopy, developed within the DIAMOND project in collaboration with the LIBELUL platform. The approach relies on a lightweight one-dimensional convolutional autoencoder trained using a self-supervised strategy, requiring neither external spectral libraries nor high signal-to-noise reference spectra. The pipeline achieves an effective workflow speedup of approximately 65× while preserving spectral fidelity and phase discrimination.

The production pipeline code is openly available on Zenodo.

Read the published article.