Publication Date
Spring 2024
Degree Type
Master's Project
Degree Name
Master of Science in Bioinformatics (MSBI)
Department
Computer Science
First Advisor
Dr. Philip Heller
Second Advisor
Dr. Thomas Austin
Third Advisor
Dr. Crawford Drury
Keywords
deep learning, convolutional neural network, Mask R-CNN, coral resilience
Abstract
As ocean temperatures rise, coral bleaching is becoming more frequent and severe. Selective breeding experiments show promise for enhancing coral resilience, but scaling these projects is hindered by the labor-intensive nature of taking numerous time series measurements as corals grow. Automating this process with computer vision is one solution to this bottleneck, and to our knowledge, no such tool exists at present. To fill this gap, we have trained a set of machine learning models, based on the Mask R-CNN framework, for segmenting juvenile corals in lab-based coral resilience research. This work shows that retraining the Mask R-CNN architecture through transfer learning results in a model capable of accurately segmenting corals in advanced developmental stages. To correctly segment images from earlier stages, an enlarged training dataset or a distinct model focused on early stages is necessary. The preliminary outcomes presented here suggest that the method holds potential.
Recommended Citation
Benbow, Jennifer, "Image Segmentation by Convolutional Neural Networks in Coral Resilience Research" (2024). Master's Projects. 1414.
https://scholarworks.sjsu.edu/etd_projects/1414
