Envisage phenotyping: integrating artificial intelligence in image analysis for selective breeding in aquaculture

Measuring metabolic and health traits in fish is essential for genetic improvement, animal welfare, and product quality in aquaculture. This thesis investigates how artificial intelligence and image analysis can be used to predict and better understand a broader range of relevant traits. It first demonstrates how image-based methods can predict production traits in seabream, while identifying the anatomical regions that contribute most to these predictions. The thesis then examines swimming performance in rainbow trout, showing that specific image-derived traits are associated with increased performance and that these relationships can be interpreted from both genetic and biological perspectives. In addition, the challenges of image-based individual identification under realistic farming conditions are evaluated. Finally, novel shape traits are developed that are objective, and heritable, for use in selection. Overall, this research shows that AI and image analysis are most effective when completely integrated within the breeding programs, with strong potential to improve animal production, health, and welfare.