<p>Small object detection in high-resolution images remains challenging because target instances usually occupy only a limited number of pixels, making their appearance cues weak and easily overwhelmed by background noise. In addition, the repeated downsampling operations used in conventional detectors tend to suppress high-frequency edge information and fine-grained spatial details, which are crucial for accurate localization of small objects. Meanwhile, many existing methods improve detection performance by introducing heavier multi-scale feature fusion strategies or more complex prediction heads, which significantly increase computational cost, parameter redundancy, and inference latency, thereby limiting real-time deployment on resource-constrained devices. To address these issues, this paper proposes GDD-YOLO, an efficient real-time small object detection framework. Specifically, a Global Edge Information Transfer (GEIT) module is designed to extract multiscale edge cues from shallow features and propagate them across the backbone to strengthen boundary-aware localization of small objects. In addition, a Dynamic Inception Mixer (DIM) is introduced to overcome the limited scale adaptability of fixed convolution kernels by performing input-adaptive multi-branch convolutional aggregation with reduced complexity. Furthermore, a lightweight detection head (DECDH) is developed using shared convolution and detail-enhanced operators to preserve contextual and local structural information while reducing parameter overhead. Experiments on the VisDrone dataset show that GDD-YOLO achieves 26.2% AP, outperforming YOLOv11-S by 2.4%, while reducing the number of parameters by 16.8% and computational cost by 14.6%. These results demonstrate that GDD-YOLO provides an effective balance between detection accuracy and inference efficiency for real-time small object detection on edge devices.</p>