Publication Date

Spring 2026

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)

Department

Computer Science

First Advisor

Philip Heller

Second Advisor

Maya deVries

Third Advisor

Genya Ishigaki

Keywords

crustose coralline algae, semantic segmentation, convolutional neural network cryptic habitat, benthic cover estimation

Abstract

Urgency surrounding coral reef restoration science is growing amid pressure from rising ocean temperatures, acidification, and habitat loss. A group of calcified red algae named crustose coralline algae (CCA) contribute to reef accretion, larval coral settlement, and structural stability. Recent surveys of reef sites in West Maui have uncovered that CCA is present in greater amounts within crevices than on exposed top-reef surfaces. Their analysis of these sites with Coral Point Count with Excel extensions (CPCe) likely underestimated the true coverage of CCA on reef surfaces. Other image processing techniques such as instance segmentation are not well-suited to this problem space. This report presents an automated, pixel-level CCA detection system based on semantic segmentation. To tackle the lack of discrete boundaries that traditional object detection models depend on we trained a convolutional neural network to perform binary classification on every pixel as belonging to CCA or not. When the trained model was applied to images previously annotated with CPCe, it delivered precise CCA image masks, supporting the hypothesis that point-count methods underestimate CCA in cryptic habitats. The results suggest that semantic segmentation offers a workable, lower-cost alternative for quantifying CCA cover, and the pipeline is designed to transfer to reef systems outside Maui.

Download

Available for download on Saturday, May 22, 2027

Share

COinS