Application of computer vision for food safety, and animal welfare and health monitoring: Application of computer vision for post-mortem and animal welfare inspection in broiler chickens and udder health monitoring in dairy goats

Abstract

Rising demand for animal-origin food and staff shortages mean that livestock farming and food production require urgent modernization. From a public health perspective, food must be safe, high-quality, and produced from animals kept under high welfare and health standards. Official veterinary supervision and food inspection are therefore essential. However, veterinary inspection faces staff shortages, subjectivity, inter-rater differences, and declining consistency during prolonged working hours. Similar challenges occur on livestock farms, where farmers cannot continuously monitor the health and welfare of hundreds or thousands of animals. Technological developments, particularly artificial intelligence (AI), may help address these challenges. Computer vision, a type of AI, enables trained models to detect abnormalities in food or animals using images or video, similarly to inspectors or farmers. This dissertation investigated two imaging techniques—computed tomography (CT) scanning and regular cameras—in two contexts: broiler carcass inspection and on-farm udder health monitoring in dairy goats. CT scanning uses X-rays to visualize internal structures non-destructively. It can examine chicken organs before their removal during slaughter and is effective in detecting bone and joint injuries. This thesis first examined limitations of current broiler carcass inspection. Computer vision models were then trained to detect wing fractures in CT scans. These fractures can indicate poor handling before slaughter and are therefore important animal welfare indicators. Another model was developed to detect unhealthy chicken livers and hearts, which are edible offal and relevant to food safety. Overall, these models could contribute to improved and automated veterinary inspection in poultry slaughterhouses. For on-farm monitoring, a computer vision model was developed to detect asymmetrical udders in dairy goats, which can indicate udder inflammation. This technology may help farmers identify diseased animals, monitor herd-level udder health, and support smart farm management systems. Overall, this work represents a step toward modernizing veterinary inspection and automating animal health and welfare monitoring. Before implementation, however, challenges involving data privacy, model ownership, liability, consumer acceptance, and potential job losses must be addressed. Further research should therefore focus not only on developing and validating AI models, but also on their societal, ethical, and economic implications.