Author Type

Graduate Student

Date of Award

Fall 12-8-2025

Document Type

Thesis

Publication Status

Version of Record

Submission Date

December 2025

Department

Computer and Electrical Engineering and Computer Science

College Granting Degree

College of Engineering and Computer Science

Department Granting Degree

Electrical Engineering and Computer Science

Degree Name

Master of Science (MS)

Thesis/Dissertation Advisor [Chair]

Xingquan Zhu

Abstract

The Florida manatee (Trichechus manatus latirostris), a threatened marine mammal often found in coastal waters, require accurate monitoring to inform conservation strategies. However, traditional counting methods are hindered by observer bias and surface distortions. Recent advances in artificial intelligence (AI) address these challenges using captured aerial footage of manatees, demonstrating the potential of artificial intelligence for automating manatee counting. Still, various challenges arise hindering accurate counts.

Building on these studies, this work advances artificial intelligence with underwater footage, instance segmentation, data augmentation, and pseudo-labeling to automate manatee counting in data-scarce environments. Three experiments were conducted using 10%, 60%, and 90% labeled training data to evaluate model performance under varying levels of data availability. Across all experiments, the proposed methods outperformed the previous state-of-the-art method, with reductions over 65% in both MAE and MASE, showing that pseudo-labeling effectively mitigates data scarcity.

Share

COinS