Gaetano Saurio

Gaetano Saurio

PhD candidate in Data Science Sapienza, University of Rome

Research & Publications

I am a Ph.D. candidate in Data Science at Sapienza University of Rome. My research focuses on Multimodal Deep Learning, Geometric Deep Learning, and Computer Vision for environmental and medical intelligence.

Google Scholar Profile ResearchGate

Current Research & Submissions (MIT Senseable City Lab)

A Multi-View Deep Learning Framework for Urban Tree Health Assessment via Fusion of Aerial Multispectral and Street-Level RGB Imagery This study addresses the monitoring of Toumeyella parvicornis in Rome. By fusing drone-based multispectral data with street-level RGB, we overcome nadir-view limitations. Benchmarking Early, Late, and Average fusion on a ResNet-18 backbone, we achieved an AUC of 0.93 and 88.9% accuracy. Status: Under Review.

Exploring the Cooling Power of Trees Using Computer Vision A large-scale analysis across Amsterdam, Boston, Dubai, and LA. We employed computer vision on thermal imagery (ADE20K segmentation) to model how genus-specific traits and water consumption influence urban cooling performance. Status: Under Review.

Peer-Reviewed Journal Articles (2025)

Micrographia in Parkinson’s Disease: Automatic Recognition through Artificial Intelligence F. Asci, G. Saurio, et al. A diagnostic tool using CNNs to detect micrographia in handwriting patterns. Read Paper

Towards rigorous dataset quality standards for deep learning tasks in precision agriculture A. Carraro, G. Saurio, F. Marinello. A framework for dataset quality metrics in agricultural AI. Read Paper

Selected Conference Papers (ICIAP)

ArcheoWeedNet: Weed Classification in the Parco archeologico del Colosseo Deep Learning system for plant recognition in archaeological contexts (>90% accuracy). URL Cite

CNNs for the Detection of Esca Disease Complex in Asymptomatic Grapevine Leaves Using 1D CNNs and Hyperspectral imaging (900–1700 nm) for early disease detection. URL Cite

PhD Research Vision: GNNs and Beyond

My research explores pushing the boundaries of Graph Neural Networks (GNNs) for plant understanding. This involves capturing long-range dependencies in complex environments and encoding robust knowledge priors into neural networks. I am currently extending these models to Multimodal Fusion and Generative AI for both plant sciences and medical diagnostics.