Simultaneous Denoising and Baseline Correction of Microplate Raman Spectra Using a Dual-Branch U-Net

Abstract

In this paper, we present a novel dual-branch U-Net architecture for the simultaneous execution of Raman baseline correction and denoising. The network features a shared encoder that diverges into two specialized decoding heads for the Raman signal and for the baseline. The two heads are coupled with a cross-attention gating mechanism. The model offers a way to cross-confirm the peaks by comparing the recovered Raman signal with the baseline corrected spectrum. Moreover, the model offers a new method for quantitative analysis by counting the overall number of photons at a deep Raman decoder block. The model was trained entirely using a custom synthetic data engine explicitly designed to emulate automated HTS acquisitions from microplates via the RamanBot platform. Comprehensive validation demonstrates robust peak recovery on synthetic spectra with signal-to-noise ratios (SNR) as low as 5. Crucially, the model successfully extracts high-fidelity signals from highly noisy glycerol and moderately noisy adenine sulfate experimental samples. Furthermore, quantitative analysis is conducted on guanine samples with different concentrations by counting the Raman photons.