Three approaches to automating taxonomy, with emphasis on the Odonata (dragonflies and damselflies)
Citation & Export
Hide
Simple citation
Kuhn, William Robert.
Three approaches to automating taxonomy, with emphasis on the Odonata (dragonflies and damselflies). Retrieved from
https://doi.org/doi:10.7282/T3SQ92Q4
Export
Description
TitleThree approaches to automating taxonomy, with emphasis on the Odonata (dragonflies and damselflies)
Date Created2016
Other Date2016-10 (degree)
Extent1 online resource (xi, 177 p. : ill.)
DescriptionTaxonomy-the field charged with naming and classifying organisms-forms a foundation for biological research. An understanding of the species on Earth is needed for informing biodiversity research, conservation efforts, management strategies, and global policy. In recent decades, a "taxonomic impediment" has arisen: there is an urgent need to know the millions of yet-undiscovered species, while funding for the science charged with this task, taxonomy, and the number of trained taxonomists are declining. This work aims to provide three software tools for taxonomists that allow them to work more efficiently and effectively, reducing this impediment. First, a system for automatically landmarking images of specimens for geometric morphometric studies was introduced, which could greatly reduce the time required to manually landmark images for these studies while also increasing the possible sample size of such studies. The system's landmarking error, however, was extremely variable on test images of the wings of dragonflies and damselflies (Odonata), and was ultimately too large (300-500 px) to compete with manual landmarking at this time. Second, a method was presented for automatically standardizing and extracting descriptive features from images of insect wings in order to quantify the appearance of the wings. The standardization method was successful in converting scans of odonate wings into consistently-formatted square images, automatically. Then, features describing the color, texture, and shape of the wings were able to be extracted, producing a small set of 663 coefficients that were able to distinguish between species. Finally, a system called Odomatic was presented and tested for automatically identifying Odonata to species from images of their wings, using the feature extraction method combined with machine learning techniques. Odomatic was able to make classifications between 32 species with expert-level (up to 92%) accuracy, making it useful for quickly identifying specimens. The tools presented here will be deployed for use by odonate researchers through the website OdonataCentral.org, but will also be released as open-sourced Python scripts so that they can be customized to be implemented on other taxonomic groups. This work will enable taxonomists and other interested parties to make easier morphological comparisons and faster identifications.
NotePh.D.
NoteIncludes bibliographical references
Noteby William Robert Kuhn
Genretheses, ETD doctoral
Languageeng
CollectionGraduate School - Newark Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.