This study uses deep learning models to monitor and evaluate cage-free hens’ welfare and behaviors. Our first project used YOLOv5 (i.e., version 5 of You Only Look At Once, an innovative deep learning model for target detection) deep learning models to track feather pecking (FP) behaviors in hens for improving early detection and minimizing the spread of FP. The YOLOv5x-pecking model performed better than YOLOv5s-pecking, achieving higher precision, recall, and mean average precision. The second project involved detecting floor eggs using YOLOv5-egg, YOLOv5x-egg, and YOLOv7-egg models. All three models detected eggs with high accuracy, while the YOLOv5x-egg model achieved the highest precision and mean average precision. The third project involved classifying the different behaviors of birds inside poultry houses using a serials of deep learning models based on YOLOv5 and YOLOv7. The YOLOv5s_BH model had the best performance in terms of precision, recall, and mean average precision compared to the other two models. These deep learning models developed in this study provide references for developing real-time automatic monitoring systems for tracking pecking damages and floor eggs and monitoring bird activity inside poultry houses to enhance welfare and reduce economic losses for egg producers.