A panting behavior-driven assessment framework for summer ventilation quality optimization in layer houses.

Ventilation quality in summer layer houses is critical for heat stress prevention, production performance, and poultry welfare. Addressing the issue of "qualified environmental parameters but chicken discomfort" caused by traditional methods overlooking spatial heterogeneity and individual differences, a dynamic ventilation quality assessment method based on panting behavior detection in laying hens was proposed. The YOLOv10-BCE panting behavior detection model was developed by embedding the BiFormer module into the backbone network to enhance multi-dimensional feature extraction, compressing neck structure parameters using the C3Ghost module, and integrating Efficient Intersection over Union (EIOU) loss to improve detection accuracy and convergence speed. K-means clustering and linear regression algorithms were employed to establish a quantitative correlation curve between ventilation quality and panting behavior, forming a Normal-Alert-Danger ventilation quality (VQ) classification standard. Experimental results demonstrated that the YOLOv10-BCE model achieved a mean average precision (mAP) of 95.8 % and a detection speed of 0.2 ms, significantly outperforming comparative models such as Faster R-CNN, SSD, and YOLOv9. The ventilation quality correlation model showed high fitting accuracy with an R² value of 0.974. Significant physiological differences (p < 0.05) in chickens across VQ grades validated the model's discriminative ability. The method accurately identified latent ventilation anomalies and spatial dead zones in large-scale layer houses. After ventilation strategy optimization, panting prevalence decreased by 65 %, establishing a closed-loop "monitoring-assessment-regulation" dynamic feedback mechanism. This study provides a behavioral-quantitative assessment solution for summer layer house ventilation quality.