DEVELOPING AND ASSESSING A DIVERSE PLANKTON IMAGERY TRAINING SET FOR MACHINE-LEARNING PLANKTON CLASSIFICATION IN THE NORTH PACIFIC SUBTROPICAL REGION - Thesis

Aug 5, 2026

Upload by Eden

The Imaging FlowCytobot (IFCB) supports oceanographic research by capturing images that reflect microbial life in the North Pacific Subtropical Gyre (NPSG). The resulting data volume makes manual sorting and taxonomic classification difficult. This thesis builds a convolutional neural network (CNN) training set to categorize IFCB images into taxonomic groups, emphasizing Hemiaulus and the Ciliphora phylum during a summer 2021 research cruise in the NPSG. Model evaluation compares automated outputs against manual annotations, tracks accuracy and precision over time, and links classifications with biovolume and particle number concentration trends. A training set of roughly 76,000 images enables robust image classification and shows improved performance despite Hemiaulus morphological changes.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages49

Reading Time2 h 3 min

AccessFree to download

Acknowledgements

Abstract

List of Tables

List of Figures

1.0 Introduction

1.1 Significance of research

1.2 The North Pacific Subtropical Gyre

1.3 Imaging FlowCytoBot

2.0 Methods

2.1 Overview

2.2 Data collection

2.3 Approach used and model development

2.4 Convolutional Neural Network

2.5 Manual vs. CNN: statistical analyses on observational data of abundances and biomass

3.0 Results

3.1 Overview

3.2 Discrepancies

4.0 Discussion

DEVELOPING AND ASSESSING A DIVERSE PLANKTON IMAGERY TRAINING SET FOR MACHINE-LEARNING PLANKTON CLASSIFICATION IN THE NORTH PACIFIC SUBTROPICAL REGION A THESIS SUBMITTED FOR PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF BACHELOR OF SCIENCE IN GLOBAL ENVIRONMENTAL SCIENCE MAY 2024 By Nicole Celine Sulla Mathews Thesis Advisors Dr. Angelicque White Dr. Fernanda Henderikx-Freitas We certify that we have read this thesis and that, in our opinion, it is satisfactory in scope and quality as a thesis for the degree of Bachelor of Science in Global Environmental Science. THESIS ADVISORS Dr. Angelicque White Department of Oceanography ________________________________ Dr. Fernanda Henderikx-Freitas Department of Oceanography ii ACKNOWLEDGEMENTS Thank you to my advisors, Angel, for introducing me to the world of microbial oceanography, and Fernanda, for always answering my countless questions and being my guide throughout this project. Thank you to the annotation team for their contributions to the training set development as well as Dr. Andrew Hirzel for overseeing the CNN development, implementations, and results used for this paper. Thank you to the White/Henderikx-Freitas Lab for providing the time, effort, and resources to support my endeavors. I would also like to thank my professors, advisors, and the Department of Oceanography for sparking my interest in this field and continually providing the opportunities to explore it. Thank you to my reviewers, who took the time and effort to help my thesis to be the best it could be. Last but certainly not least, si yu'os ma'åse' tomy family – for always being there for support, inspiration, and guidance – you are the reason I am where I am today. ABSTRACT The Imaging FlowCytobot (IFCB) has a continually growing role in oceanographic research, particularly in the exploration of microbial life within the North Pacific Subtropical Gyre (NPSG). However, the vast amount of data generated by the IFCB poses a challenge for manual sorting and taxonomic classification. This study addresses this challenge by developing a Convolutional Neural Network (CNN) training set to efficiently categorize IFCB images into taxonomic groups. Specifically focusing on the diatom Hemiaulus and ciliate phylum Ciliphora during a research cruise within the NPSG in the summer of 2021, the study aims to quantify the CNN's performance compared to manual annotations ofIFCB images taken on this cruise, providing insights into the CNN’s accuracy and precision over time. Statistical analyses of the CNN’s machine learning-based classifications indicate a high accuracy in the automated identification of Hemiaulus and Ciliophora. Analysis of biovolume and particle number concentration reveals trends in taxonomic abundance over the course of the cruise. Despite morphological changes of Hemiaulus as it loses structure over time, the CNN demonstrates an overall improvement in accuracy as the cruise progresses, particularly for Hemiaulus. This study highlights the development of a robust training set of roughly 76,000 images, allowing the CNN to accurately classify images collected within the NPSG. Keywords: Taxonomic sorting, machine learning, zooplankton, Imaging FlowCytoBot (IFCB), North Pacific Subtropical Gyre (NPSG), ocean microbiology. TABLE OF CONTENTS Acknowledgements ......................................................................................................................... iii Abstract .............................................................................................................................. iv [List of Tables ..................................................................................................................... vi](List of Tables ..................................................................................................................... vi) List of Figures .............................................

FAQs

What problem does the study address in IFCB oceanographic research?

The study addresses the challenge of manually sorting and taxonomically classifying the large volumes of data generated by the Imaging FlowCytobot (IFCB).

Which taxa and setting are used to train and evaluate the CNN?

The training and evaluation focus on the diatom Hemiaulus and the ciliate phylum Ciliphora, using IFCB images collected during a research cruise in the North Pacific Subtropical Gyre (NPSG) in summer 2021.

How is CNN performance assessed over the course of the cruise?

Performance is quantified by comparing CNN classifications to manual annotations and by analyzing accuracy and precision trends over time, alongside changes in morphology and associated abundance patterns.

Aug 5, 2026

Upload by Eden