Intelligent Poultry Health Recognition Using an Improved YOLOv8

Intelligent poultry health recognition is crucial in precision livestock farming. Nevertheless, existing methods suffer from limited precision and practicality due to complex backgrounds and small detection targets. This study aims to develop an improved YOLOv8-based poultry health recognition model to enhance detection precision and robustness in real poultry farming environments. In this study, YOLOv8 is improved at both the data and network levels. At the data level, multi-strategy data augmentation is applied. At the network level, the neck network of YOLOv8 is reconstructed by integrating the large-small (LS) block to enhance multi-scale feature fusion capabilities. Experimental results indicate that both improvement levels increase precision by 4.4% and 4.5%, respectively. Moreover, the model achieves overall improvements of 5.4% in precision and 9.9% in mAP@0.5:0.95 compared with the original YOLOv8 algorithm. In this study, the developed model effectively improves poultry health recognition performance and demonstrates practical value for practical poultry farming applications.