Deep learning in poultry farming: comparative analysis of Yolov8, Yolov9, Yolov10, and Yolov11 for dead chickens detection.

Abstract

Automated detection of dead chickens is essential for enhancing biosecurity, animal welfare, and operational efficiency in poultry farms. This study evaluates the performance of YOLOv8n, YOLOv9c, YOLOv10n, and YOLOv11n for detecting dead chickens in cage-free poultry farms. A synthetic dataset of 3413 images was created by compositing manually annotated images of dead and healthy chickens into realistic stall backgrounds to simulate real farm conditions. The models were assessed using standard object detection metrics (precision, recall, and mean average precision (mAP) at IoU thresholds of 0.5 and 0.5-0.95) alongside computational efficiency indicators including inference speed, frames per second (FPS), model size, and training time. YOLOv9c achieved the highest detection accuracy (mAP@50 = 0.983, mAP@50-95 = 0.93), making it the most reliable for minimising false positives and missed detections. YOLOv11n delivered the fastest inference speed (2.8 ms/frame, ∼357 FPS), making it more suitable for real-time applications. These results underscore the importance of selecting a YOLO model based on farm-specific operational constraints. YOLOv9c is recommended for accuracy-critical tasks, YOLOv11n for real-time monitoring, and YOLOv8n or YOLOv10n for resource-limited deployments. Comparative analysis with earlier YOLO models (YOLOv3-YOLOv7) shows that newer versions improve both detection reliability and processing speed. This work contributes a performance benchmark to guide AI-based poultry monitoring and highlights future directions, including real-world deployment and validation under live farm conditions.