Back to search

University of Southampton

Computer vision and explainable AI for morphological interpretation: applications in taxonomy, cryptic species, and natural history collections

Abstract

dc:description.abstract

Taxonomy provides the foundation for ecology, evolution, and conservation, but it continues to face significant challenges. Morphology-based approaches are hampered by issues such as cryptic similarity, phenotypic plasticity, and the global decline in taxonomic expertise. Molecular methods, while precise in detecting genetic divergence, often fail to identify the visible traits that separate lineages. This persistent disconnect between genotype and phenotype limits the integration of molecular and morphological evidence in biodiversity research. The rapid digitisation of natural history collections now provides unprecedented image datasets, but their scale demands automated, interpretable approaches for morphological analysis. This thesis investigates the application of computer vision and explainable artificial intelligence to the interpretation of morphology across species, populations, and collections. The central aim is to evaluate whether computer vision models, when coupled with heatmap-based interpretability methods, can both achieve high classification accuracy and reveal biologically meaningful traits in ways transparent to human experts.<br/>Four empirical studies demonstrate this potential. Chapter 2 trained convolutional neural networks to classify cryptic limpets from the Baja California peninsula, achieving accuracies marginally higher than expert taxonomists. Heatmaps confirmed that models attended to diagnostic shell features. Chapter 3 developed a computer vision pipeline to detect mislabelled butterfly specimens in the digitised Lepidoptera collection of the Natural History Museum, London, revealing many potential errors where many were verified by experts and genetics. Chapter 4 applied computer vision to genetically divergent but morphologically similar limpet clades, showing that saliency maps and shape analyses could expose subtle, previously overlooked shell differences aligned with phylogeographic structure. Chapter 5 compared human and machine attention in the classification of British butterflies, demonstrating substantial overlap between heatmap focus and diagnostic traits in identification guides, while also highlighting features not mentioned in the literature and attention to novel artefacts. Together, these studies show that explainable computer vision can complement and extend traditional taxonomy: accelerating large-scale identifications, flagging curation errors, and uncovering cryptic morphological divergence. More broadly, they position interpretable artificial intelligence as a practical instrument for biodiversity science, one that can bridge molecular and morphological evidence, strengthen the curation of natural history collections, and provide reproducible baselines for the study and conservation of biodiversity.<br/>

Degree

thesis:*
Name dc:type.qualificationname
Ph.D.
Level dc:type.qualificationlevel
doctoral
Grantor dc:publisher.institution
University of Southampton
Year dc:date.issued
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Hollister, Jack Daniel
Advisors dc:contributor.advisor
  • Fenberg, Phillip
  • Horton, Tammy
  • Cai, Xiaohao
  • Price, Ben
ORCID person:identifier.orcid

Rights

Language dc:language
English, eng

Chain of custody

source
Harvested from
University of Southampton
Base URL
eprints.soton.ac.uk/cgi/oai2
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Hollister, Jack Daniel. Computer vision and explainable AI for morphological interpretation: applications in taxonomy, cryptic species, and natural history collections. doctoral thesis, University of Southampton, 2025.