Modelling the relationship of exterior traits of fish
In aquaculture, accurate real-time data on the health and performance of fish are essential for stabilizing product quality and selecting the best fish in breeding programs. Collecting these data is costly, timeconsuming, and stressful for the animals. Health and performance are affected by stress, hormones, energy, fat metabolism and diseases. Modern breeding programs measure these traits because they impact economically important production traits such as growth, survival and feed conversion rate (FCR). Traditional measurements of health and metabolic state often require handling the animals or taking invasive physical samples that require effort, money, and time. Consequently, phenotypic measures lag the real-time development of the animal. Alternatively, image analysis may offer a fast, non-invasive and labour reducing alternative. Previous studies have identified a link between fish metabolism and external phenotypes but assessing the health and metabolic performance of fish using image analysis requires a direct link between these traits and features of the image. In this project, we propose to investigate imageextracted features as health and metabolic indicators and to develop predictive models for health and metabolic status using real-time imaging. The project will combine appropriate fish images, cutting-edge image analysis, and health condition and metabolic data as inputs for machine learning and deep learning with artificial intelligence as decision support for breeding programs. The output is expected to reveal how metabolism and health conditions interact and result in external phenotypes. The resulting models will be applicable in high-throughput phenotyping for metabolic phenotype prediction, disease detection and welfare assessment.