Computer vision in wildlife disease surveillance

Olsen, Agnethe 2025. Computer vision in wildlife disease surveillance. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF (PhD thesis) - Accepted Post-Print Version Download (44MB) | Preview
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (171kB)

Abstract

Wildlife disease surveillance is often limited by the scale, cost, and logistical challenges of traditional methods. This thesis aimed to investigate the potential of using computer vision to classify visible signs of disease in wildlife from digital imagery. In Chapter 1, we introduced the challenges of traditional surveillance and the potential for computer vision, while in Chapter 2 , we provided an overview of the precedents for using computer vision in related fields. In Chapter 3, we analysed the WOAH animal disease database and found that over two-thirds (67%) of major infectious diseases present with externally visible signs, and that diseases spread by direct contact were significantly more likely to have them, confirming a broad scope for image-based monitoring. Chapter 4 then investigated wildlife imagery availability in online repositories, using salmonids as a case study. Our analysis of nearly 70,000 images showed that platforms like iNaturalist are a vast and growing data source, and that a consistent 14-18% of images displayed signs of disease or damage, highlighting an underused resource for health monitoring. In Chapter 5, we developed deep learning models to classify Saprolegnia spp. infection in salmonids. The best model achieved a high macro-average F1-score (0.930), with performance strongly influenced by dataset composition, as taxonomically focused and balanced datasets yielded the best results. Chapter 6 used rumpwear in common brushtail possums as a second case study to explore disease severity assessment. We showed a model's output probability for the 'Disease' class served as a robust, well-calibrated proxy for severity, distinguishing between mild and obvious signs, and found that semi-supervised learning provided minimal benefit. Finally, Chapter 7 discusses how these results show computer vision is a viable tool to complement traditional surveillance, providing a framework to understand disease dynamics and support more timely and effective conservation responses.

Item Type:	Thesis (PhD)
Date Type:	Published Online
Status:	Unpublished
Schools:	Schools > Biosciences
Subjects:	Q Science > Q Science (General)
Date of First Compliant Deposit:	12 February 2026
Last Modified:	12 Feb 2026 15:08
URI:	https://orca.cardiff.ac.uk/id/eprint/184748

Actions (repository staff only)

Edit Item