Evaluating cleaning efficacy by image-based machine learning: case study of soot removal from silk
Year: 2026
Authors: Cremonesi M., Pastorelli G., Yang N., Al-Emam E., Martinelli E., Comelli D., Dal Fovo A., Fontana R., Cicchi R., Saez NO., Berwouts J., Janssens K., van der Snickt G.
Autors Affiliation: Univ Antwerp, Dept Design Sci, Antwerp Cultural Heritage Sci ARCHES Res Grp, Antwerp, Belgium; Univ Antwerp, Antwerp X ray Imaging & Spect Lab AXIS Res Grp, Antwerp, Belgium; Natl Gallery Denmark Statens Museum Kunst SMK, Copenhagen, Denmark; Univ Ghent, Fac Engn & Architecture, Dept Appl Phys, Res Unit Plasma Technol RUPT, Ghent, Belgium; Sohag Univ, Fac Archaeol, Dept Conservat, Sohag, Egypt; Politecn Milan, Dept Phys, Milan, Italy; INO Natl Inst Opt, CNR, Florence, Italy; European Lab Nonlinear Spect, Sesto Fiorentino, Italy; Univ Antwerp, StatUa Ctr Stat, Antwerp, Belgium.
Abstract: In the context of the assessment and validation of plasma-generated atomic oxygen (AO), within the framework of EU Horizon MOXY project, this study compared three non-invasive methodologies for the semi-quantitative evaluation of cleaning efficacy: CIELAB colorimetry, statistical analysis of image brightness histograms and a supervised machine learning method (TWS). The research focuses on assessing the strengths and limitations of these methods when applied to complex, highly textured substrates such as textiles. A benchmark set of simplified model systems (SMSs), consisting of pongee silk swatches artificially soiled with soot and treated with AO alongside seven alternative cleaning methods, was used as case study. Among the evaluated techniques, the machine-learning-based approach demonstrated high reliability and versatility for the selective detection of heterogeneous soiling on highly reflective surfaces. Due to its open-access design and user-friendly interface, the method has strong potential for wider use in systematically evaluating cleaning efficacy across various conservation contexts.
Journal/Review: NPJ HERITAGE SCIENCE
Volume: 14 (1) Pages from: 393-1 to: 393-14
More Information: This research was carried out and funded by the EU Horizon (Topic ID: HORIZON-CL2-2021-HERITAGE-01-01, Green technologies and materials for cultural heritage) MOXY project (grant no. 101061336) and FWO research project Plasmart (grant G0C2822N). Special thanks to Bruce Banks and Sharon K. Rutledge (NASA) for the original proof of concept of the plasma-generated atomic oxygen technology. Thanks to Dr. Jan Dariusz Cutajar for the prompt help and support.KeyWords: AlgorithmDOI: 10.1038/s40494-026-02717-y
