Abstract: To address main challenges faced by robots when detecting, locating, and picking up dead hens in intelligent caged laying hen farms, such as complex backgrounds, low detection accuracy, and difficulties in target localization, a method based on an improved YOLOv8n model for detecting dead hens in caged laying hen farms and locating their pickup regions has been proposed, combined with a depth camera for localization．Firstly, to improve detection accuracy, a Shuffle Attention mechanism has been added before SPPF layer in backbone network to enhance model's feature extraction capability．Additionally, C2f-Faster module has been introduced to replace C2f, further strengthened spatial feature extraction by reducing redundant computation and memory access．After completing target detection at pickup regions, model for locating dead hens was optimized by integrating a depth camera（Realsense D435i）and employing multi-sampling point rubust depth measurement, and adaptive coordinate smoothing based on target．Experimental results have shown that improved YOLOv8n detection model achieved a precision of 90.1%, a recall rate of 90.6%, and mean average precision of 93.2%, with 6.0% reduction in number of parameters, 14.9% reduction in computational load, and 1.5% increase in inference speed compared to original model．Improved YOLOv8n model demonstrated high-precision recognition and localization of dead hens pickup regions even under challenging conditions of high density, occlusion, and varying environmental factors, indicating exceptional robustness．