The occurrence of abnormal eggs is often closely related to the health status of poultry, and egg detection can promptly identify health issues in chickens while improving overall egg quality. To address the challenges of egg adhesion and occlusion between eggs and wire mesh in chicken coops, a lightweight object detection model named ATW – YOLO was proposed for normal/ abnormal egg detection. Firstly, part of the standard convolutions in the backbone and neck networks were replaced with ADown modules to achieve model lightweighting and extract locally salient features. Then, the triplet attention mechanism was incorporated into the C2PSA module to enhance the model??s ability to extract cross-dimensional features, effectively mitigating egg adhesion and occlusion issues. Additionally, a weighted bidirectional feature pyramid network (BiFPN) was introduced to strengthen feature fusion performance and improve model accuracy. Finally, the WIoU loss function was adopted to increase the model's focus on mediumquality samples. Experimental results demonstrated that the ATW – YOLO lightweight object detection model achieved precision (P), recall (R), and mean average precision (mAP0.5) of 90.4% , 89.5% , and 93.3% , respectively. Compared with the YOLO 11n model, these metrics showed improvements of 1.9, 2.5, and 2.0 percentage points, respectively. The model parameters and floating-point operations (FLOPs) were 5.2 × 10? and 2.18 × 10?, representing reductions of 4 × 10? and 1. 1 × 10? compared with that of YOLO 11n. The detection speed reached 197.2 frames per second ( FPS ), meeting the requirements for real-time egg recognition. In conclusion, the ATW – YOLO model exhibited superior comprehensive performance in normal/ abnormal egg detection tasks, providing technical support for automated and intelligent management in poultry farming.