Purpose:

Surveying coral reefs is traditionally a manual, labor-intensive process conducted by divers. This limits the frequency and breadth of data collection, leaving gaps during critical environmental events such as heatwaves or hurricanes. To address this challenge, we partner with the University of Puerto Rico to deploy low-cost surface vessels equipped with camera systems to map reefs visually.

I specifically focused on automatic coral reef topic modeling to identify potentially sick or dying coral and assess overall reef system health in a more efficient, automated way. I implemented the Segment Anything Model (SAM) and K-Means clustering techniques to segment and analyze coral reef images. By clustering extracted features, we derive topics representing different coral landscapes—enabling us to quickly detect health anomalies and monitor reef conditions at scale.

Repository Link