Edge AI and IoT-enabled livestock intrusion detection and automated farm protection using YOLOv8s and MQTT-enabled event notification

Precision livestock farming relies on intelligent sensing and automated decision-making to reduce economic losses from livestock intrusion. Existing systems typically handle visual detection or IoT notification separately, limiting real-time autonomous response. This paper presents an Edge AI and IoT framework that integrates YOLOv8s-based cattle detection with MQTT-driven event communication and automated physical actuation. The system combines a real-time object detection engine, Python-based orchestration, MQTT communication, an ESP32 controller, cloud notification services, and acoustic deterrence into a unified cyber-physical architecture. The model was trained on a Roboflow cattle dataset using transfer learning under Ultralytics YOLOv8. Experimental results show a Precision of 0.8039, Recall of 0.6846, F1-score of 0.7394, mAP@50 of 0.8094, and mAP@50:95 of 0.5347. Average inference latency was 132.16 ms, while MQTT transmission required only 0.21 ms. This study demonstrates the feasibility of integrating visual perception, IoT communication, automated notification, and real-time physical intervention for autonomous livestock intrusion monitoring.