Development of deep learning technologies for behaviour and health monitoring in Precision Livestock Farming

The use of artificial intelligence technologies in animal agriculture is transforming the way livestock systems are managed and understood. This dissertation explores how computer vision and deep learning can be applied in the context of Precision Livestock Farming to monitor animal behaviour, health, and welfare. The main intention is to show how digital innovation may balance production goals with the ethical and environmental responsibilities associated with contemporary farming. The research combines expertise from computer science, animal behaviour, and sustainability studies, and is articulated through two main case studies. The first focuses on dairy cattle, where deep learning models were trained on real farm data to identify individual cows. The results confirm that computer vision can provide alternatives to traditional monitoring approaches. The second case study extends these methods to horses, developing models for posture classification and keypointbased pose estimation. Complementary investigations extend the scope of the thesis beyond technical innovation. A Life Cycle Assessment evaluates the environmental impacts of PLF technologies in dairy production, while a review on heat stress in cattle considers animal welfare within the context of climate change. These studies demonstrate the transformative potential of computer vision to support smarter, more ethical, and more sustainable livestock systems.