Honors College Thesis
 

Analysis of Convolutional Neural Network Image Pre-processing Methods

Public Deposited

Contenu téléchargeable

Télécharger le fichier PDF
https://ir.library.oregonstate.edu/concern/honors_college_theses/sf268d34d

Descriptions

Attribute NameValues
Creator
Abstract
  • Oceanic plankton have both a large impact on our oceans health and atmo- spheric balance of Co2. The overall health of planktonic life is determined by many physical, chemical, and ecological factors, that drive taxonomic abundances and the relative amount of non-living biomass called detritus. Recent advances in microscopic imaging allow censusing of thousands of particles, plankton or detritus, at high spatial or temporal resolution, which necessitates machine learning models to rapidly clas- sify images into appropriate categories. Convolutional Neural Networks fulfill these criteria for aiding in the predictions of oceanic health and diversity. However, little has been researched to improve these models performance by processing input before model training. This study focuses on image pre-processing techniques for optimiz- ing the accuracy of classifying detritus and non-detritus. Findings suggest that less processed input trained on by the model results in less overfitting, but little to no cor- relation with higher accuracy or higher F1-score. Similarly, the more dense the input trained on by the model, the faster the model overfits.
Resource Type
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Non-Academic Affiliation
Déclaration de droits
Publisher
Peer Reviewed
Language

Des relations

Parents:

This work has no parents.

Dans Collection:

Articles

La vignette Titre Date de téléchargement Visibilité actes
file details JacobsenMatthew2022.pdf 2022-08-29 Public Télécharger