Year: 2021
Type: dissertation
Abstract: Zooplankton have been used as indicators of aquatic ecosystem health, but their identification using traditional microscopy can be time consuming. Machinery such as Yokogawa Fluid Imaging Technologies FlowCam allow for rapid imaging of microscopic particles, such as zooplankton. However, these machineries lack sophisticated algorithms like convolutional neural networks (CNNs) to classify images. Here, the efficacy of CNNs to classify six freshwater zooplankton species and a fiber-like particle with images obtained from a FlowCam Macro were examined. Oxford Visual Geometry Group (VGG) CNN architecture best suited our data. CNNs were found to have validation accuracies greater than 97%. The models were further tested with independent random mixtures of species and fiber-like particles, and they achieved accuracies ranging from 84.9% to 96.33% and F1-scores ranging from 82.43% to 95.02%. These results indicate that CNNs are promising tools for the rapid identification of zooplankton species.
Source: TSpace (University of Toronto)
Author: Corey D Ramkissoon
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funders: Ministry of Environment, Mitacs
OpenAlex ID: W3140362864