Airport Bird Detection and Recognition Based on YOLOv11n

Abstract

Bird strike is a typical sudden-onset disaster endangering civil aviation safety. Accurate and efficient bird target detection serves as a core prerequisite for airports to prevent bird strike incidents. To tackle the challenges of tiny bird targets in complex airport environments, including low pixel coverage, weak feature information and severe background interference, this paper presents AeroBird-YOLO, a lightweight improved YOLOv11n algorithm for airport bird detection.In the backbone network, lightweight spatial-to-depth convolution (L-SPDConv) is introduced in shallow layers, while a C3k2-DCNv4 module is constructed in deep layers to adapt to the non-rigid deformation of birds via the dynamic sampling mechanism of fourth-generation deformable convolution. Omni-dimensional dynamic convolution (ODConv) is embedded in the neck feature fusion stage to suppress background noise from airport runways and other facilities, and the Normalized Wasserstein Distance (NWD) loss function is adopted at the prediction head.Ablation experiments and multi-algorithm comparisons on the re-annotated and augmented AirBirds dataset demonstrate that AeroBird-YOLO achieves a mean average precision (mAP@50) of 91.71%, representing a 4.09 percentage point improvement over the baseline YOLOv11n model