Chen, Yueh-ShaoYueh-ShaoChenRustia, Dan Jeric ArcegaDan Jeric ArcegaRustiaHuang, Shao-ZhengShao-ZhengHuangJIH-TAY HSUTA-TE LIN2025-09-112025-09-112025-09https://www.scopus.com/record/display.uri?eid=2-s2.0-105007710607&origin=resultslisthttps://scholars.lib.ntu.edu.tw/handle/123456789/731985Article number: 101674This study presents an IoT-enabled cow face recognition system leveraging edge computing to enable real-time, automated monitoring of individual cow feeding behavior. The system integrates a lightweight YOLOv4-tiny model for cow face detection with MobileNetV2 for feature extraction, optimized for embedded devices with limited computational power. A key innovation is the incorporation of few-shot learning (FSL), allowing the system to adapt efficiently to newly introduced cows with minimal training data. The algorithm achieved robust performance, with an F1-score of 0.98 for detection and a recognition accuracy of 0.97 using FSL. Feeding times estimated by the system were validated against manually observed data, demonstrating high precision with a mean absolute error (MAE) of 1.7 min per cow. Long-term experiments conducted under varying seasonal conditions showcased the system's effectiveness in monitoring feeding behavior year-round. Results revealed significant seasonal differences, with cows feeding longer in winter (197.0 min/day) than in summer (115.5 min/day), likely due to the effects of heat stress during warmer months. This IoT-driven system offers scalable, non-invasive monitoring solutions for dairy farm environments, enabling real-time insights to support herd management, early health issue detection, and individualized feeding strategies. By integrating advanced IoT technologies with agricultural practices, this system provides a pathway to enhancing productivity and animal welfare in precision dairy farming.falseEdge computingEmbedded systemFace recognitionFew-shot learningLivestock monitoring[SDGs]SDG13journal article10.1016/j.iot.2025.1016742-s2.0-105007710607