AquaGuard: AI-Assisted Underwater Marine Debris Detection and Reporting — A Literature Review and System Design

Underwater marine debris is hard to monitor because observations are gathered with different tools, judged by different people and stored in formats that cannot be compared. Deep-learning detectors have been applied to sonar, remotely operated vehicle (ROV) and towed-camera imagery, yet the work reviewed here concentrates on detection models and annotated datasets rather than on the workflow that surrounds a detection. This paper surveys research on marine-debris detection and monitoring, compares existing approaches by sensing modality, detector family, data and workflow support, and derives six research gaps. It then presents AquaGuard, a modular system in which a single-stage object detector sits behind a Flask backend with role-based access, a SQLite store, reviewer-in-the-loop correction and shared web and Flutter mobile front ends. The contribution is a design argued from the literature: no detection accuracy is reported, and the held-out evaluation that must precede any performance claim is specified. Development status and open risks, including domain shift and false positives from natural background, are discussed.