Abstract
This study developed and evaluated an IoT-enabled smart incubator integrated with a YOLO-based computer vision system for the non-invasive detection of chicken egg fertility. Traditional candling, although widely practiced in small- scale poultry operations, remains labor-intensive, subjective and prone to inconsistent interpretation. To address these limitations, the proposed system combines automated fertility classification, real-time environmental monitoring and closed-loop incubation control within a unified platform. The hardware architecture employs an Arduino Mega 2560 for actuator management and a Raspberry Pi 4 for image processing, sensor integration, local data handling and cloud synchronization. A PID controller was implemented to regulate incubation temperature and humidity, while system data were transmitted to Firebase for remote monitoring through a mobile application. For model evaluation, three YOLOv8 variants (Nano, Small and Medium) were trained and tested using 4,500 annotated candling images collected on incubation Days 1, 6, 12 and 18. Experimental results showed that YOLOv8s achieved the best overall performance, obtaining an mAP@0.5 of 0.943, precision of 1.00 and recall of 0.99, while YOLOv8n delivered comparable accuracy with lower computational complexity, indicating strong suitability for edge deployment. The incubator’s control system maintained a mean temperature of 37.5°C (SD = 0.12°C), automatically increased humidity to 85.5% during hatching and recorded an average end-to-end latency of 0.43 seconds. User evaluation further indicated high usability, demonstrating that the proposed system is an effective, practical and scalable solution for improving hatchery monitoring and fertility assessment in small- to medium-scale poultry production.
Keywords: Computer Vision, Egg Fertility Detection, IoT Incubator, nYOLOv8, PID Control, Poultry Automation.