<![CDATA[The increasing demand for sustainable, welfare-oriented livestock management necessitates innovative solutions for behavior monitoring, particularly in group-housed settings, where challenges such as animal density and overlapping bodies hinder traditional observation methods. This study introduces a Convolutional Neural Network (CNN)-based model enhanced with segmentation techniques to accurately classify behaviors among group-housed pigs, a context in which individual monitoring is crucial for welfare assessment, disease prevention, and production efficiency. By leveraging segmentation, the model isolates individual pigs in video footage, overcoming occlusion issues and significantly improving classification accuracy. This approach not only advances the analysis of animal behavior in dense environments but also aligns with the principles of innovation, promoting the adoption of AI-driven monitoring solutions in livestock management. In comparison with various models, YOLOv11m-augmentation achieved the highest mAP@0.5 score of 0.969 and a notable precision of 0.925. This CNN and segmentation-based method effectively identifies key behaviors, including eating, drinking, sleeping, and standing, with particularly high precision for behaviors most indicative of animal welfare. This research contributes to sustainable livestock practices by offering a scalable, cost-effective technology for real-time welfare assessment, potentially reducing labor requirements, enhancing farm management decisions, and promoting animal health. The study’s findings underscore the potential of integrating innovation principles with AI in agriculture, presenting a viable pathway toward sustainable livestock management practices that balance productivity with animal welfare.]]>
Copyrights © 2025
