A Computer Vision Method for Monitoring Early Symptoms of Respiratory Diseases in Leghorn Laying Hens

Infectious Laryngotracheitis (ILT), caused by an alphaherpesvirus (ILTV), is an acute respiratory disease that inflicts significant economic and welfare impacts on poultry through increased mortality, reduced weight gain, and declining egg production. Current surveillance depends on clinical observation followed by labor-intensive laboratory diagnostics, delaying early intervention. Clinical and behavioral signs, including eye rubbing, head bobbing, mouth breathing, huddling, and reduced feeding and drinking, typically precede severe outcomes, making automated early detection critical for disease control. This pilot study deployed vision-based object detection to identify ILT-associated signs, comparing YOLO11-obb and YOLO26-obb models. We divided 14 SPF Leghorn hens into ILTV-infected (n = 7) and uninfected control (n = 7) groups. We recorded high-definition videos over 8 h daily for seven days post-infection, yielding 1214 annotated images split into training (70%), validation (20%), and testing (10%) sets. Models were evaluated using precision, recall, mAP@0.50, and mAP@0.50–0.95. YOLO26x-obb achieved the highest overall performance, while feeding and drinking behaviors were detected with a precision of 0.93 to 0.98 across both model variants. Detection of clinical signs ranged from 0.36 to 1.0, reflecting visual overlap between behaviors. These findings establish a scalable, non-invasive framework for real-time ILT monitoring, enabling earlier management intervention before flock-level disease spread.