Title here
Summary here
Raman Denoising Pipeline
Artificial intelligence is core to DIAMOND’s expertise for the PEPR DIADEM community. This workflow illustrates that expertise applied to experimental characterization: a practical Noise2Noise deep learning pipeline for denoising high-throughput Raman spectroscopy data, developed within the DIAMOND project in collaboration with the LIBELUL platform. It relies on a lightweight one-dimensional convolutional autoencoder trained using a self-supervised deep learning 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 method and its validation are described in:
- A Practical Noise2Noise Denoising Pipeline for High-Throughput Raman Spectroscopy, Advanced Engineering Materials (2026)
The pipeline code is openly available on:
The raw Raman spectra used to train and evaluate the pipeline are openly available on Zenodo.