Document Type

Honors Thesis

Major

Computer Engineering

Advisor(s)

Vikas Dhiman

Committee Members

Prabuddha Chakraborty, Melissa Ladenheim, Gregory Studer

Graduation Year

December 2024

Publication Date

Spring 5-2024

Abstract

A Remotely Operated Vehicle was designed, constructed, and programmed as a senior design project in Electrical and Computer Engineering by Dyllon Dunton and Jacob Wildes. The system was intended to be an alternative means to inspect the underside of ships. Given the small footprint of the system, it can be easily extended into other applications. In this thesis project, the observation system is modified to detect if a fish is present or not, classify the species, localize where in the image the fish is, and mask the fish by separating fish pixels from non-fish pixels. Additionally, the original design will be analyzed and compared with the new system. Finally, existing issues with both systems and next steps are discussed. Open-source solutions were systematically searched, and several candidate solutions were found. The candidates were then evaluated in terms of ease of use and similarity to the goals of this project. Eventually, custom solution using the YOLOv8 backbone, and a custom labeled dataset was developed which achieved a precision of 86.5%, and a mAP-50 of 91%.

Download

Share