DOSA-YOLO: Improved Model Research for the Detection of Common Chicken Diseases Using Phenotypic Features

Abstract

Chicken farming plays a crucial role in the global food supply; however, the frequent occurrence of chicken diseases presents a substantial challenge to the industry’s sustainable development. This study introduces an enhanced YOLOv11 model, DOSA-YOLO, designed to detect four prevalent chicken diseases: avian pox, coccidiosis, Mycoplasma gallisepticum, and Newcastle disease. The research team developed an intelligent inspection robot to capture multi-angle images in intensive farming environments, constructing a five-class dataset comprising 8052 images. These images were categorized based on phenotypic features such as comb, eyes, and wattles, as well as pathological anatomical characteristics. To address challenges such as complex backgrounds, multi-scale lesions, and occlusion interference, three attention-enhancement modules—MSDA, MDJA, and SEAM—were integrated into the YOLOv11. The model was trained and validated using the constructed dataset and compared against seven other algorithms, including YOLOv5n, YOLOv7tiny, YOLOv8n, YOLOv9t, YOLOv10n, YOLOv11n, YOLOv12n, and Faster R-CNN. Experimental results demonstrated that DOSA-YOLO achieved a mean Average Precision (mAP) of 97.2% and an F1-score of 95.0%, outperforming the seven other algorithms while maintaining a balance between lightweight design and performance with GFLOPs of 6.9 and 2.87 M parameters. The model provides strong support for real-time chicken health monitoring in intensive farming environments.