Deep Learning-Assisted Dynamic Mode Decomposition (DA-DMD)

🔬 Overview

DA-DMD is a physics-informed framework currently established for removing non-resonant background (NRB) in Coherent Anti-Stokes Raman Spectroscopy (CARS) using:

• Unsupervised DMD: Hankelized Dynamic Mode Decomposition with 2-class clustering
• DA-DMD: Deep Learning with channel attention + CNN (Adaptive weighting of DMD modes for robust Raman signal recovery)

Methods for NRB Removal.

📄 Paper

Our method is described in:
Deep Learning-Assisted Dynamic Mode Decomposition for Non-resonant Background Removal in CARS Spectroscopy
Proceeding at DAGM German Conference on Pattern Recognition (GCPR 2025)
📥 Download PDF (postprint)

💻 Code

Implementation available here:
👉 GitHub Repository

👥 Authors

Adithya Ashok Chalain Valapil, Carl Messerschmidt, Maha Shadaydeh, Michael Schmitt,
Jürgen Popp, Joachim Denzler

BibTeX:

@InProceedings{valapil2025dadmd,
author="Chalain Valapil, Adithya Ashok
and Messerschmidt, Carl
and Shadaydeh, Maha
and Schmitt, Michael
and Popp, J{\"u}rgen
and Denzler, Joachim",
title="Deep Learning-Assisted Dynamic Mode Decomposition for Non-resonant Background Removal in CARS Spectroscopy",
booktitle="Pattern Recognition (DAGM GCPR 2025)",
series="Lecture Notes in Computer Science",
year="2026",
publisher="Springer Nature Switzerland",
address="Cham",
pages="41--56",
isbn="978-3-032-12840-9",
doi="https://doi.org/10.1007/978-3-032-12840-9_4"
}

🙏 Acknowledgements

This work is funded by the European Union’s Horizon Europe research and innovation program under Grant Agreement no. 101135175.

For more works visit Computer Vision Group Jena.