AI toolkit for phenotyping bird populations
For NIOO-KNAW (The Netherlands Institute of Ecology), we developed a computer vision and AI toolkit for phenotyping bird populations. This software toolkit can handle complex datasets, containing both bird images and information from field work and other sources. An important part of the toolkit is a vision algorithm that can segment an image of a bird’s head into its constituent parts. Once we know which pixels belong to the bill, eye ring, iris and pupil, we can perform a detailed and accurate analysis on their shape, color profile and texture. The ecologists at NIOO plan to use this information to automate and improve their research. Assessing the visual appearance of birds automatically can boost our understanding of genotype-phenotype relations and the effects that environment and diet have on the birds. The toolkit can also be used to perform automatic gender classification and age estimation. Using this toolkit as an example, we are now looking into broader applicability.
This project was performed in collaboration with the Van de Pol Group at NIOO-KNAW, and Sovon Vogelonderzoek Nederland. It was financed by the KNAW 3V Fund. The development of this toolkit took place after our first bird phenotyping project with Judith Risse at NIOO-KNAW.