This study addresses the critical challenge of moving object detection in airport surveillance systems, where extreme scale variations and class imbalance significantly impact detection performance. We introduce AeroYOLO, an advanced object detection model tailored for airport environments, which extends YOLOv11 by incorporating three key architectural enhancements: a Swin Transformer backbone for better small-object perception, an Adaptive Receptive Field module for scale-aware detection, and Optimal Transport Assignment to mitigate class imbalance issues. Additionally, we propose a Dynamic Scale Augmentation training strategy that improves model robustness to object scale variations. To support research in this domain, we present the Airport Surface Moving Object Dataset, a new benchmark comprising 1,240 airplane, 1,455 vehicle, and 734 person instances captured from real-world airport operations. Experimental results demonstrate that AeroYOLO achieves superior performance in detecting objects of varying scales in complex airport environments, particularly improving the detection of small, safety-critical objects such as ground personnel and service vehicles. Furthermore, to validate the generalization capability of our approach, we also evaluate AeroYOLO on the public remote sensing dataset NWPU VHR-10, where it continues to demonstrate strong detection performance.