Wavenumber-Aware Deep Learning for Robust Analysis of Raman Spectra
Abstract
Machine-learning models for spectroscopy usually treat a spectrum as a fixed-grid intensity vector, leaving the physical coordinate implicit in the array index. Such models are tied to a single sampling grid, spectral range and resolution, which limits their use when measurement conditions change. Here we represent each spectral point as a pair of intensity and wavenumber and compute Transformer positional encodings from the wavenumber itself. On Raman spectra of three bacterial species and their synthetic mixtures, this representation remained robust to noise, sparser sampling and partial spectral coverage across convolutional, residual and Transformer networks. In controlled comparisons, wavenumber-based encoding outperformed index-based and learned encodings under partial observation and retained its performance on sampling grids never seen during training, where index-based encoding degraded. These findings were reproduced on a public 30-isolate bacterial benchmark. Explicit physical coordinates offer a simple route to robust spectral analysis under variable measurement conditions.
Related
- Construction and Evaluation of a Ketosis Prediction Model Based on Milk MIR Spectra and Feature Wavenumber Selection in Dairy Cows.
- Deposit optics, film thickness and acquisition date confound Raman spectra of dried droplets
- Fast and accurate simulation of Raman spectra of gold-organic systems
- Fast and accurate simulation of Raman spectra of gold–organic systems
- Syntheses, X‐ray Crystal Structures, and Raman Spectra of Self‐Assembled XeO3 Chain and Network Complexes of Polytopic sp2‐ and sp3‐Nitrogen Bases
BibTeX
@misc{kim2026wavenumberaware,
title = {{Wavenumber-Aware Deep Learning for Robust Analysis of Raman Spectra}},
author = {Jihwan Kim and Jaeseong Heo and Yujin Kim and Sang Bin Yoon and Chanjoong Kim and Kwang-Im Oh and Hyon Bin Na and Kyungwon Kwak and Myeongkee Park and Joonyoung F. Joung},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.30746},
eprint = {2610.30746},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.30746},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - UNPB
TI - Wavenumber-Aware Deep Learning for Robust Analysis of Raman Spectra
AU - Jihwan Kim
AU - Jaeseong Heo
AU - Yujin Kim
AU - Sang Bin Yoon
AU - Chanjoong Kim
AU - Kwang-Im Oh
AU - Hyon Bin Na
AU - Kyungwon Kwak
AU - Myeongkee Park
AU - Joonyoung F. Joung
PY - 2026
DA - 2026/10/22
PB - arXiv
DO - 10.48550/arxiv.2610.30746
UR - https://arxiv.org/abs/2610.30746
AB - Machine-learning models for spectroscopy usually treat a spectrum as a fixed-grid intensity vector, leaving the physical coordinate implicit in the array index. Such models are tied to a single sampling grid, spectral range and resolution, which limits their use when measurement conditions change. Here we represent each spectral point as a pair of intensity and wavenumber and compute Transformer positional encodings from the wavenumber itself. On Raman spectra of three bacterial species and their synthetic mixtures, this representation remained robust to noise, sparser sampling and partial spectral coverage across convolutional, residual and Transformer networks. In controlled comparisons, wavenumber-based encoding outperformed index-based and learned encodings under partial observation and retained its performance on sampling grids never seen during training, where index-based encoding degraded. These findings were reproduced on a public 30-isolate bacterial benchmark. Explicit physical coordinates offer a simple route to robust spectral analysis under variable measurement conditions.
ER -