Ultra-high resolution imagery and AI for avian flu impact assessment

We are collaborating with the Scottish Seabird Centre and colleagues within the School of Biological Sciences to deploy the latest imaging and machine learning technologies in support of research into the ongoing avian flu epidemic affecting seabirds across the northern hemisphere.

Our work on the Bass Rock includes support for an MSc dissertation entitled 'Quantifying the Impact of Avian Influenza on the Northern Gannet Colony of Bass Rock using Ultra-High Resolution UAV Imagery and Deep Learning' by Dr Amy Tyndall, which received national TV coverage on the BBC Breakfast program and national news bulletins. BBC Article 2023.

Following the outbreak of avian flu on the Bass Rock in 2022, the Scottish Seabird Center published an urgent request for support to conduct an aerial survey over the colony to assess the immediate impact of the epidemic. We were happy to help and within a few days we were able to conduct the first drone-based colony-wide survey on the island using our DJI Matrice 300 RTK drone with L1 sensor. Acquiring a large number of overlapping images across the island we were then able to use photogrammetric software to create a single composite orthomosaic image of the colony. In 2023 we returned with our new P1 ultra-high resolution camera on the Matrice 300 and repeated the survey at even higher resolution.

Using the resulting orthomosaic images, Amy used machine learning techniques to automatically identify gannets within the images and provide a count of the colony population at the time of each survey. Amy was also able to train the software to distinguish between live birds, dead birds and flying birds, based on the distinctive body morphology associated with each.

These initial studies have also led to our involvement a major NERC Pushing the Frontiers grant to investigate the avian flu outbreak (led by Dr Emma Cunningham, School of Biological Sciences) and a NERC Discipline Hopping grant to assess drone sensing techniques for identifying individual birds with a population.