Automated Welfare Assessment of Dairy Cattle Using Artificial Intelligence

Sammanfattning: The recent boom in Artificial Intelligence (AI) and intelligent decision-support systems has revolutionized efficiency across industries, and there is great value in applying these findings to the dairy industry. With the growth of the dairy industry comes a need for new and innovative tools to support the increased demand. AI can help dairy farmers achieve considerable improvements in the success of their business and assist in more ethical farming practices where animal welfare is a top priority. The earlier we are able to accurately detect medical ailments, the better for the quality of life, the success of the business, and the overall decision-making process. This master’s thesis project explores data collection methods and applies state-of-the-art machine learning models and practices to determine the feasibility of building cost-effective and accurate systems for monitoring animal welfare. We analyze the required granularity of data required to achieve a desired predictive capability, propose the architecture of one such system and evaluate its accuracy. We show that an early detection system can achieve a mean average precision of 30% and above at a 50% intersection over union threshold. However, executing such a system on edge devices creates additional difficulties and enforces higher requirements on the image capture quality due to the lower frame rate capabilities. This project shows the challenges and opportunities in building such a system.