The School of Computer Science is pleased to present…
Scene Graph Generation for Coastal Monitoring
MSc Thesis Defense by:
Nathan Cherry
Date: April 17th, 2026
Time: 11:00AM
Location: 122 Essex Hall
Abstract:
Efficient coastal management requires a transition from basic image collection to structured, semantic understanding of shoreline dynamics. This research addresses the high cost of traditional remote sensing by introducing a novel framework that applies Scene Graph Generation (SGG) to the coastal domain. By bridging the gap between raw citizen-science imagery and actionable intelligence, the work provides a scalable methodology for digitizing the logical structure of diverse coastal environments.
The core of this thesis is the Coastal Monitoring Scene Graph Dataset (CMSGD), a specialized benchmark derived from the Coastie Initiative. Built on a domain-specific taxonomy, CMSGD captures intricate geomorphological features and ecological relationships with a relational density significantly higher than standard benchmarks. Results demonstrate that models fine-tuned on CMSGD vastly outperform general SGG models, establishing stablish a proof-of-concept for the feasibility of an automated, explainable system to assist researchers and policymakers in tracking shoreline evolution.
Thesis Committee:
Internal Reader: Dr. Boubakeur Boufama
External Reader: Dr. Mohammad Hassanzadeh
Advisor: Dr. Ziad Kobti & Dr. Chris Houser
Chair: Dr. Shaon Shuvo
