Denoising U-Net models for automated estimation of the colorimetric sensor responses

A denoising deep learning architecture based on the U-Net was developed for automated estimation of the colorimetric sensor response, accounting for realistic conditions such as variable perspectives, sensor shear deformation, rotation, and non-uniform illumination across the sensor surface. The U-Net-based module denoises the image and corrects illumination artefacts, while a multi-layer perceptron (MLP) module is jointly trained to do regression of the target parameter values based on the corrected latent representation of the sensor. Reference:Wireless Colorimetric Multi-Biomarker Sensing to Enable Critical Neonatal Monitoring.Alejandra Castelblanco, Elisabetta Ruggeri, Giusy Matzeu, Motaharehsadat Heydarian, Kai Foerster, Ahmed Bahnasy, Andreas Flemmer, Julia A. Schnabel, Benjamin Schubert, Fiorenzo G. Omenetto, Anne HilgendorffmedRxiv 2025.09.21.25336258; doi: https://doi.org/10.1101/2025.09.21.25336258