An improved YOLOv8s algorithm has been proposed to address challenges encountered in object detection within airfield environments, such as significant variations in image sizes and substantial occlusion of targets. These issues often result in reduced model accuracy. The enhancement involves integrating multiple attention mechanisms into the YOLOv8s model. The first improvement involves upgrading the C2f module by introducing a Receptive-Field Attention Convolution mechanism into the backbone. This mechanism adapts the network to different input features by assigning specific weights to each receptive field. Additionally, a Multi-Scale Dilated Attention mechanism has been introduced, partitioning the feature map into various heads and applying different dilation rates for feature extraction at multiple scales. This enhancement aims to improve the multi-scale feature extraction capability. Furthermore, improvements have been made to the object detection head using Dynamic Bottle Block (DBB). This improvement enhances the network’s feature representation by integrating convolutions of different sizes and configurations. Experimental results indicate that the improved YOLOv8s model has increased mean Average Precision (mAP) metrics by 1.9% and 1.5% for mAP50 and mAP50:95, respectively, on the Key-Airports dataset. These enhancements effectively enhance object detection capabilities in airfield environments.