Graduate Degree Type
Computer Information Systems (M.S.)
School of Computing and Information Systems
The goal of this qualitative research project is to develop and optimize a multi-class discrimination model to identify different species of coral based on their digital images. Currently, there are artificial intelligence (AI) models that can distinguish between coral and other undersea objects such as sand or rocks, but to our knowledge the problem of multi-species classification has not yet been addressed. Given that coral reefs are a good indicator of overall ocean health, it is important to develop models that can classify the presence of different species in underwater images as a way to monitor the effects of climate change.
The dataset for this project consists of images of various species of coral; collected from the reef regions of offshore Florida and Bonaire, sanitized, labeled, and organized according to species. This study explores multiple options for image pre-processing, compares different model architectures, and experiments with hyperparameters such as learning rate with a goal of developing the most accurate coral species classifier. Our preliminary results: using only a portion of the complete dataset, a multi-class coral species classifier was produced that achieves 92.2% accuracy.
Jang, Hyeong Gyu, "Building a Deep Model for Multi-class Coral Species Discrimination" (2022). Culminating Experience Projects. 223.