Cattle Anomaly Behavior Detection System Based on IoT and Computer Vision in Precision Livestock Farming

Efficient livestock management is crucial to maintain food security and economic stability. One of the main challenges in this sector is identifying abnormal livestock behavior, which can affect their health, productivity, and overall well-being. This paper proposes an IoT and computer vision-based solution for the real-time monitoring of cattle behavior. The system combines CCTV cameras, Raspberry Pi, and the YOLOv8 detection model to capture and analyze cattle activities. The data were processed and presented on a web interface, allowing farmers to take prompt action. The initial findings indicate that the system achieves over 88% accuracy in detecting behaviors such as standing, eating, and lying down. However, challenges arise when identifying cattle with different body colors or partially obstructed areas. The Difference and Isolation Forest methods were employed to address anomalies using thresholds derived from the field data. These methods offer early indications of potential stress in cattle. Continuous 24/7 monitoring enables daily behavioral analysis, with cameras strategically positioned to capture data with a detection loss of 10.7% to 16.57%. The experiment also uncovered abnormal activities leading up to the Eid al-Adha celebration in Indonesia when the farm experienced unusually high traffic. This result demonstrates the system’s ability to detect behavioral shifts under non-ideal conditions. Thus, the proposed solution has the potential for broader implementation, offering enhancements in livestock welfare and farm productivity with further expansion of data variation.