Unveiling Renaissance Drawing Techniques: A Multimodal Machine Learning Approach to the Analysis of Giulio Romano’s Amazzonomachia

Claudia Scatigno, Silvia Giampaolo, Gabriella Pace, Serena Galetti, Maura Picciau, Giulia Festa,
Unveiling Renaissance Drawing Techniques: A Multimodal Machine Learning Approach to the Analysis of Giulio Romano’s Amazzonomachia",
Journal of Molecular Structure,
2025,
144614,
ISSN 0022-2860,
https://doi.org/10.1016/j.molstruc.2025.144614.
(https://www.sciencedirect.com/science/article/pii/S0022286025032570)
Abstract: During the Italian Renaissance, drawings were primarily created as preparatory studies for final works, including sculptures, paintings, and frescoes. Today, these drawings not only illustrate the conceptual and technical processes employed by artists but also enhance our understanding of the materials and techniques of the period. The analysis of such drawings involves execution methods such as strokes, line drawing, shading, and hatching, while also offering valuable insight into the development of art prints during that period. In this context, advanced technologies, such as spectroscopy and machine learning (ML) have been successfully employed to access the state of conservation, raw materials and execution methods of these masterpieces. Spectroscopy enables the identification of pigments, inks, and papers, while machine learning models can enhance the interpretation of complex data, revealing deeper insights into artistic processes, material degradation, and the evolution of drawing techniques. Here, an innovative machine learning approach is developed for the analysis of composition, degradation, and artistic techniques of an Italian Renaissance drawing attributed to Giulio Romano, titled Amazzonomachia (16th century) and preserved at the Istituto Centrale per la Grafica in Rome. By integrating multimodal spectroscopic data (e.g. X-ray fluorescence spectroscopy - XRF, Fourier transform infrared spectroscopy - FTIR) with machine learning models, we identify the use of lead-based mine as the preferred workshop drawing material, and the use of iron gall ink to emphasize line strokes, providing valuable insights into the preservation and restoration of these masterpieces. The results demonstrate a robust multimodal framework for analyzing historical drawing by integrating spectroscopy and machine learning, establishing a potential new standard in the field. The approach bridges material science and heritage studies, pushing cultural heritage diagnostics into the AI era.
Keywords: spectral signatures; machine learning in art drawing conservation; spectroscopy